sync from monorepo @ 2452e92e

crates/dirigent_anth/CLAUDE.md

@@ -0,0 +1,148 @@
+# Package: dirigent_anth
+
+Claude Code JSONL session parser and toolkit.
+
+## Quick Facts
+- **Type**: Library
+- **Main Entry**: src/lib.rs
+- **Dependencies**: serde, serde_json, chrono, uuid, camino, thiserror, tracing, dirs
+- **Status**: Core parsing complete — ready for downstream consumers
+
+## Purpose
+
+Reads Claude Code's local JSONL session storage (`~/.claude/projects/`) and produces typed, deduplicated, correlated Rust data structures. The types are the product — downstream consumers (archivist import, shell usage analyzers, session browsers) depend on these structs.
+
+## Key Features
+
+- **Session Discovery**: Scan `~/.claude/projects/` for all Claude Code projects and sessions
+- **JSONL Parsing**: Lenient line-by-line parser that handles unknown fields and message types
+- **Streaming Dedup**: Collapse streamed assistant messages to their final version
+- **Tool Correlation**: ID-based pairing of tool_use → tool_result across parallel calls
+- **Conversation Tree**: Reconstruct uuid/parentUuid threading with branch detection
+- **Noise Classification**: Identify meta messages, warmup, interruptions, API errors
+- **Sub-Agent Loading**: Recursive parsing of sub-agent JSONL with metadata
+- **Timestamp Parsing**: Handle ISO 8601, Unix seconds, and Unix milliseconds
+
+## Architecture
+
+### Design Principles
+
+1. **Types are the product** — Well-typed Rust structs that downstream consumers import
+2. **Lenient parsing** — Unknown fields ignored, unknown message types logged and skipped
+3. **Stream-oriented** — Line-by-line BufReader parsing, never loads entire files
+4. **Sync-first** — File parsing is CPU-bound; no async overhead
+5. **Cross-platform** — camino::Utf8PathBuf throughout for Windows/Unix compatibility
+
+### Module Organization
+
+- **`types.rs`** — All public data types (Content, ContentBlock, RawMessage variants, ToolCall, etc.)
+- **`error.rs`** — AntError enum with I/O, JSON parse, home-not-found, invalid-path variants
+- **`parser.rs`** — JSONL line parser and file parser with lenient error handling
+- **`dedup.rs`** — Streaming deduplication of assistant messages by uuid
+- **`correlation.rs`** — Tool call ↔ result pairing by tool_use_id
+- **`tree.rs`** — Conversation tree from uuid/parentUuid relationships
+- **`noise.rs`** — Noise pattern classification (meta, warmup, interruptions, etc.)
+- **`discovery.rs`** — Filesystem scanning for Claude projects and sessions
+- **`subagent.rs`** — Sub-agent JSONL and metadata loading
+- **`util.rs`** — Timestamp parsing utilities
+
+## Public API
+
+### Quick Start
+
+```rust
+use dirigent_anth::{discover_claude_home, discover_projects, load_session};
+
+// Discover all projects
+let home = discover_claude_home()?;
+let projects = discover_projects(&home)?;
+
+// Load a session with full parsing
+for project in &projects {
+    for session_ref in &project.sessions {
+        let session = load_session(session_ref)?;
+        println!("Messages: {}, Tools: {}, Subagents: {}",
+            session.messages.len(),
+            session.tool_exchanges.len(),
+            session.subagents.len());
+    }
+}
+```
+
+### Key Functions
+
+| Function | Purpose |
+|----------|---------|
+| `discover_claude_home()` | Find `~/.claude/` directory |
+| `discover_projects(home)` | Scan for all project directories |
+| `parse_session(path)` | Parse a JSONL file into messages |
+| `parse_session_deduped(path)` | Parse with streaming dedup applied |
+| `dedup_messages(msgs)` | Deduplicate streamed assistant messages |
+| `correlate_tools(msgs)` | Pair tool calls with results by ID |
+| `ConversationTree::build(msgs)` | Build conversation tree |
+| `classify_noise(msg)` | Classify a message as noise |
+| `load_subagents(dir)` | Load sub-agent sessions from artifacts |
+| `load_session(ref)` | Full parse: dedup + correlate + tree + subagents |
+| `parse_timestamp(value)` | Parse ISO/Unix timestamps |
+
+## Data Model
+
+### Claude Code JSONL Format
+
+Each line in `~/.claude/projects/<encoded-path>/<session-uuid>.jsonl` is a JSON object with a `type` field discriminator. Five types: `user`, `assistant`, `progress`, `system`, `queue-operation`.
+
+- **Outer wrapper**: camelCase fields (sessionId, parentUuid, isSidechain, gitBranch)
+- **Inner message body**: snake_case fields (stop_reason, tool_use_id, is_error)
+- **Content**: Either a plain string or array of typed content blocks
+
+### Content Blocks
+
+| Type | Fields |
+|------|--------|
+| text | `text` |
+| tool_use | `id`, `name`, `input` |
+| tool_result | `tool_use_id`, `content`, `is_error` |
+| thinking | `thinking` |
+| image | `source` |
+
+Unknown content block types are silently dropped (lenient deserialization).
+
+## Testing
+
+```bash
+cargo test --package dirigent_anth
+```
+
+Tests use synthetic JSONL fixtures in `tests/fixtures/`:
+- `minimal_session.jsonl` — Basic session with all message types
+- `streaming_dedup.jsonl` — Streaming dedup scenario
+- `tool_correlation.jsonl` — Parallel and sequential tool calls
+- `branching_tree.jsonl` — Conversation with branches
+- `noise_patterns.jsonl` — All noise pattern types
+- `subagent/` — Sub-agent session with parent and metadata
+
+## Error Handling
+
+- Individual unparseable JSONL lines are logged and skipped (lenient)
+- I/O errors and missing directories are propagated as AntError
+- Unknown message types are skipped via serde
+- Unknown content blocks are silently filtered
+
+## Related Packages
+
+- **dirigent_archivist** — Future consumer for session import
+- No current dependencies on other dirigent packages (standalone)
+
+## Future Enhancements
+
+- Bash command analysis module (shell usage analytics)
+- Archivist event transform/import
+- CLI tool with scan/analyze/import subcommands
+- SQLite caching layer
+- Watch mode for new session monitoring
+
+## Documentation
+
+- **Package README**: `./README.md` - User-facing overview
+- **API Docs**: Run `cargo doc --package dirigent_anth --open`
+- **Design Plan**: `docs/superpowers/plans/2026-03-23-dirigent-ant-design.md`

crates/dirigent_anth/Cargo.toml

@@ -0,0 +1,37 @@
+[package]
+name = "dirigent_anth"
+version = "0.1.0"
+edition = "2021"
+
+[lib]
+path = "src/lib.rs"
+
+[[bin]]
+name = "anth_bear"
+path = "src/bin/anth.rs"
+
+[[bin]]
+name = "anth_usage"
+path = "src/bin/anth_usage.rs"
+
+[features]
+default = []
+dirigent-paths = ["dep:dirigent_config"]
+
+[dependencies]
+serde = { version = "1.0", features = ["derive"] }
+serde_json = "1.0"
+chrono = { version = "0.4", features = ["serde"] }
+chrono-tz = "0.10"
+uuid = { version = "1.11", features = ["serde"] }
+camino = { version = "1.1", features = ["serde1"] }
+dirs = "6.0"
+thiserror = "2.0"
+tracing = "0.1"
+regex = "1"
+portable-pty = "0.8"
+vt100 = "0.15"
+dirigent_config = { path = "../dirigent_config", optional = true }
+
+[dev-dependencies]
+tempfile = "3.0"

crates/dirigent_anth/src/anth_usage.rs

@@ -0,0 +1,331 @@
+use chrono::{Datelike, NaiveDate, NaiveTime, Utc};
+use chrono_tz::Tz;
+use serde::Serialize;
+
+#[derive(Debug, Serialize, Default)]
+pub struct UsageData {
+    pub gauges: Vec<UsageGauge>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub contributions: Option<ContributionInfo>,
+}
+
+#[derive(Debug, Serialize)]
+pub struct UsageGauge {
+    pub name: String,
+    pub percent_used: u32,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub resets: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub resets_iso: Option<String>,
+}
+
+#[derive(Debug, Serialize, Default)]
+pub struct ContributionInfo {
+    #[serde(skip_serializing_if = "Vec::is_empty")]
+    pub factors: Vec<ContributionFactor>,
+    #[serde(skip_serializing_if = "Vec::is_empty")]
+    pub subagents: Vec<SubagentUsage>,
+}
+
+#[derive(Debug, Serialize)]
+pub struct ContributionFactor {
+    pub description: String,
+    pub percent: u32,
+}
+
+#[derive(Debug, Serialize)]
+pub struct SubagentUsage {
+    pub name: String,
+    pub percent: u32,
+}
+
+pub struct ProcessedOutput {
+    pub raw_screen: String,
+    pub data: UsageData,
+}
+
+pub fn process_usage_screen(raw: &str) -> ProcessedOutput {
+    let lines: Vec<&str> = raw.lines().collect();
+
+    let start = lines
+        .iter()
+        .position(|l| {
+            let t = l.trim();
+            t.starts_with('─') && t.chars().filter(|&c| c == '─').count() >= 6
+        })
+        .unwrap_or(0);
+
+    let end = lines
+        .iter()
+        .rposition(|l| !l.trim().is_empty())
+        .map(|i| i + 1)
+        .unwrap_or(lines.len());
+
+    let clean_lines = &lines[start..end];
+    let raw_screen = clean_lines.join("\n");
+
+    let data = extract_usage_data(clean_lines);
+
+    ProcessedOutput { raw_screen, data }
+}
+
+fn extract_usage_data(lines: &[&str]) -> UsageData {
+    let mut data = UsageData::default();
+    let mut i = 0;
+
+    while i < lines.len() {
+        let trimmed = lines[i].trim();
+
+        if (trimmed.starts_with("Current session") || trimmed.starts_with("Current week"))
+            && !trimmed.contains('%')
+        {
+            let name = trimmed.to_string();
+            if let Some(gauge) = find_gauge(&lines[i..], &name) {
+                data.gauges.push(gauge);
+            }
+        }
+
+        if let Some(factor) = parse_contribution_factor(trimmed) {
+            data.contributions
+                .get_or_insert_with(ContributionInfo::default)
+                .factors
+                .push(factor);
+        }
+
+        if trimmed.starts_with("Subagents") {
+            let subs = parse_subagent_table(&lines[i + 1..]);
+            if !subs.is_empty() {
+                data.contributions
+                    .get_or_insert_with(ContributionInfo::default)
+                    .subagents = subs;
+            }
+        }
+
+        i += 1;
+    }
+
+    data
+}
+
+fn find_gauge(lines: &[&str], name: &str) -> Option<UsageGauge> {
+    let mut percent = None;
+    let mut resets_raw = None;
+
+    for line in lines.iter().skip(1).take(4) {
+        let t = line.trim();
+        if let Some(pct) = extract_percent_used(t) {
+            percent = Some(pct);
+        }
+        if t.starts_with("Resets ") {
+            resets_raw = Some(t.trim_start_matches("Resets ").to_string());
+        }
+    }
+
+    percent.map(|p| {
+        let resets_iso = resets_raw.as_deref().and_then(parse_reset_to_iso);
+        UsageGauge {
+            name: name.to_string(),
+            percent_used: p,
+            resets: resets_raw,
+            resets_iso,
+        }
+    })
+}
+
+/// Parse reset strings like:
+///   "12:30pm (Europe/Vienna)"          → today at 12:30 in that tz
+///   "May 12, 9am (Europe/Vienna)"      → May 12 at 09:00
+///   "May 12, 9:30am (Europe/Vienna)"   → May 12 at 09:30
+///   "Jun 1, 12pm (America/New_York)"   → Jun 1 at 12:00
+///
+/// Claude Code uses JS `Intl.DateTimeFormat` style output.
+fn parse_reset_to_iso(s: &str) -> Option<String> {
+    // Split off the timezone from parentheses
+    let (datetime_part, tz_str) = {
+        let open = s.rfind('(')?;
+        let close = s.rfind(')')?;
+        let tz = s[open + 1..close].trim();
+        let dt = s[..open].trim();
+        (dt, tz)
+    };
+
+    let tz: Tz = tz_str.parse().ok()?;
+    let now = Utc::now().with_timezone(&tz);
+
+    let (date, time_str) = if datetime_part.contains(',') {
+        // "May 12, 9am" or "May 12, 9:30am"
+        let comma_pos = datetime_part.find(',')?;
+        let date_part = datetime_part[..comma_pos].trim();
+        let time_part = datetime_part[comma_pos + 1..].trim();
+
+        let date = parse_month_day(date_part, now.year())?;
+        (date, time_part)
+    } else {
+        // "12:30pm" — today in the given timezone
+        (now.date_naive(), datetime_part)
+    };
+
+    let time = parse_12h_time(time_str)?;
+    let naive = date.and_time(time);
+    let local = naive.and_local_timezone(tz).earliest()?;
+    let utc = local.with_timezone(&Utc);
+
+    Some(utc.to_rfc3339())
+}
+
+/// Parse "May 12", "Jun 1", "December 25", etc.
+fn parse_month_day(s: &str, year: i32) -> Option<NaiveDate> {
+    let parts: Vec<&str> = s.split_whitespace().collect();
+    if parts.len() != 2 {
+        return None;
+    }
+    let month = match parts[0].to_lowercase().as_str() {
+        "jan" | "january" => 1,
+        "feb" | "february" => 2,
+        "mar" | "march" => 3,
+        "apr" | "april" => 4,
+        "may" => 5,
+        "jun" | "june" => 6,
+        "jul" | "july" => 7,
+        "aug" | "august" => 8,
+        "sep" | "september" => 9,
+        "oct" | "october" => 10,
+        "nov" | "november" => 11,
+        "dec" | "december" => 12,
+        _ => return None,
+    };
+    let day: u32 = parts[1].parse().ok()?;
+    NaiveDate::from_ymd_opt(year, month, day)
+}
+
+/// Parse "9am", "12pm", "9:30am", "12:30pm"
+fn parse_12h_time(s: &str) -> Option<NaiveTime> {
+    let s = s.trim().to_lowercase();
+    let is_pm = s.ends_with("pm");
+    let is_am = s.ends_with("am");
+    if !is_pm && !is_am {
+        return None;
+    }
+
+    let num_part = &s[..s.len() - 2];
+
+    let (hour, minute) = if let Some((h, m)) = num_part.split_once(':') {
+        (h.parse::<u32>().ok()?, m.parse::<u32>().ok()?)
+    } else {
+        (num_part.parse::<u32>().ok()?, 0)
+    };
+
+    let hour_24 = match (hour, is_pm) {
+        (12, true) => 12,
+        (12, false) => 0,
+        (h, true) => h + 12,
+        (h, false) => h,
+    };
+
+    NaiveTime::from_hms_opt(hour_24, minute, 0)
+}
+
+fn extract_percent_used(line: &str) -> Option<u32> {
+    let line = line.trim();
+    if !line.ends_with("% used") {
+        return None;
+    }
+    let before_pct = line.trim_end_matches("% used").trim();
+    before_pct
+        .rsplit_once(char::is_whitespace)
+        .map(|(_, n)| n)
+        .unwrap_or(before_pct)
+        .parse()
+        .ok()
+}
+
+fn parse_contribution_factor(line: &str) -> Option<ContributionFactor> {
+    if !line.contains("% of your usage") {
+        return None;
+    }
+    let pct_str = line.split('%').next()?;
+    let percent: u32 = pct_str.trim().parse().ok()?;
+    let description = line.to_string();
+    Some(ContributionFactor {
+        description,
+        percent,
+    })
+}
+
+fn parse_subagent_table(lines: &[&str]) -> Vec<SubagentUsage> {
+    let mut subs = Vec::new();
+    for line in lines {
+        let t = line.trim();
+        if t.is_empty() || t.starts_with('─') || t.contains("to day") || t.contains("to cancel") {
+            break;
+        }
+        if let Some(pos) = t.rfind('%') {
+            let num_start = t[..pos]
+                .rfind(char::is_whitespace)
+                .map(|i| i + 1)
+                .unwrap_or(0);
+            if let Ok(pct) = t[num_start..pos].parse::<u32>() {
+                let name = t[..num_start].trim().to_string();
+                if !name.is_empty() && !name.contains("% of") {
+                    subs.push(SubagentUsage {
+                        name,
+                        percent: pct,
+                    });
+                }
+            }
+        }
+    }
+    subs
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn parse_time_only() {
+        let t = parse_12h_time("12:30pm").unwrap();
+        assert_eq!(t, NaiveTime::from_hms_opt(12, 30, 0).unwrap());
+    }
+
+    #[test]
+    fn parse_time_am() {
+        let t = parse_12h_time("9am").unwrap();
+        assert_eq!(t, NaiveTime::from_hms_opt(9, 0, 0).unwrap());
+    }
+
+    #[test]
+    fn parse_time_12am() {
+        let t = parse_12h_time("12am").unwrap();
+        assert_eq!(t, NaiveTime::from_hms_opt(0, 0, 0).unwrap());
+    }
+
+    #[test]
+    fn parse_time_with_minutes() {
+        let t = parse_12h_time("9:30am").unwrap();
+        assert_eq!(t, NaiveTime::from_hms_opt(9, 30, 0).unwrap());
+    }
+
+    #[test]
+    fn parse_reset_time_only() {
+        let iso = parse_reset_to_iso("12:30pm (Europe/Vienna)");
+        assert!(iso.is_some());
+        let iso = iso.unwrap();
+        assert!(iso.contains("T"));
+        // Should end in +00:00 (UTC via rfc3339)
+        assert!(iso.ends_with("+00:00"));
+    }
+
+    #[test]
+    fn parse_reset_date_and_time() {
+        let iso = parse_reset_to_iso("May 12, 9am (Europe/Vienna)").unwrap();
+        assert!(iso.contains("T07:00:00") || iso.contains("T08:00:00"));
+        // CEST is UTC+2, CET is UTC+1 — depends on whether May 12 is summer time
+    }
+
+    #[test]
+    fn parse_month_day_basic() {
+        let d = parse_month_day("May 12", 2026).unwrap();
+        assert_eq!(d, NaiveDate::from_ymd_opt(2026, 5, 12).unwrap());
+    }
+}

crates/dirigent_anth/src/bin/anth.rs

@@ -0,0 +1,252 @@
+//! Minimal CLI for dirigent_anth — validate parsing and search sessions.
+//!
+//! Usage:
+//!   cargo run --package dirigent_anth --bin ant             # validate all sessions
+//!   cargo run --package dirigent_anth --bin ant -- search "query"  # search user messages
+//!   cargo run --package dirigent_anth --bin ant -- stats    # show statistics
+
+use dirigent_anth::*;
+use std::io::BufRead;
+
+fn main() {
+    let args: Vec<String> = std::env::args().skip(1).collect();
+
+    let home = match discover_claude_home() {
+        Ok(h) => h,
+        Err(e) => {
+            eprintln!("Could not find Claude home: {e}");
+            std::process::exit(1);
+        }
+    };
+
+    let projects = match discover_projects(&home) {
+        Ok(p) => p,
+        Err(e) => {
+            eprintln!("Could not discover projects: {e}");
+            std::process::exit(1);
+        }
+    };
+
+    match args.first().map(|s| s.as_str()) {
+        Some("search") => {
+            let query = args.get(1).map(|s| s.as_str()).unwrap_or("");
+            if query.is_empty() {
+                eprintln!("Usage: ant search <query>");
+                std::process::exit(1);
+            }
+            cmd_search(&projects, query);
+        }
+        Some("stats") => cmd_stats(&projects),
+        Some("validate") | None => cmd_validate(&projects),
+        Some(other) => {
+            eprintln!("Unknown command: {other}");
+            eprintln!("Commands: validate (default), search <query>, stats");
+            std::process::exit(1);
+        }
+    }
+}
+
+/// Validate that the parser can handle all sessions without errors.
+fn cmd_validate(projects: &[ClaudeProject]) {
+    let mut total_sessions = 0;
+    let mut total_ok = 0;
+    let mut total_messages = 0;
+    let mut total_skipped_lines = 0;
+    let mut errors: Vec<(String, String)> = Vec::new();
+
+    for project in projects {
+        println!(
+            "Project: {} ({} sessions)",
+            project.original_path,
+            project.sessions.len()
+        );
+
+        for session in &project.sessions {
+            total_sessions += 1;
+
+            // Raw line-level validation: count how many lines parse vs skip
+            let (_raw_ok, raw_skip) = validate_lines(&session.jsonl_path);
+            total_skipped_lines += raw_skip;
+
+            // Full pipeline validation
+            match load_session(session) {
+                Ok(parsed) => {
+                    total_ok += 1;
+                    total_messages += parsed.messages.len();
+                    let tools = parsed.tool_exchanges.len();
+                    let subs = parsed.subagents.len();
+                    let branches = if parsed.tree.is_linear() {
+                        "linear"
+                    } else {
+                        "branched"
+                    };
+
+                    if raw_skip > 0 {
+                        println!(
+                            "  {} — {} msgs, {} tools, {} subagents, {} | {raw_skip} lines skipped",
+                            &session.id[..8.min(session.id.len())],
+                            parsed.messages.len(),
+                            tools,
+                            subs,
+                            branches,
+                        );
+                    }
+                }
+                Err(e) => {
+                    errors.push((session.id.clone(), e.to_string()));
+                    eprintln!("  {} — ERROR: {e}", &session.id[..8.min(session.id.len())]);
+                }
+            }
+        }
+    }
+
+    println!("\n--- Validation Summary ---");
+    println!("Projects:  {}", projects.len());
+    println!("Sessions:  {total_sessions} ({total_ok} ok, {} errors)", errors.len());
+    println!("Messages:  {total_messages}");
+    if total_skipped_lines > 0 {
+        println!("Skipped:   {total_skipped_lines} unparseable lines");
+    }
+
+    if !errors.is_empty() {
+        println!("\nErrors:");
+        for (id, err) in &errors {
+            println!("  {id}: {err}");
+        }
+        std::process::exit(1);
+    }
+}
+
+/// Count parseable vs skipped lines in a JSONL file.
+fn validate_lines(path: &camino::Utf8Path) -> (usize, usize) {
+    let file = match std::fs::File::open(path.as_std_path()) {
+        Ok(f) => f,
+        Err(_) => return (0, 0),
+    };
+    let reader = std::io::BufReader::new(file);
+    let mut ok = 0;
+    let mut skip = 0;
+
+    for (i, line) in reader.lines().enumerate() {
+        let line = match line {
+            Ok(l) => l,
+            Err(_) => {
+                skip += 1;
+                continue;
+            }
+        };
+        if line.trim().is_empty() {
+            continue;
+        }
+        if parse_line(&line, i + 1).is_some() {
+            ok += 1;
+        } else {
+            skip += 1;
+        }
+    }
+
+    (ok, skip)
+}
+
+/// Search user messages for a query string (case-insensitive).
+fn cmd_search(projects: &[ClaudeProject], query: &str) {
+    let query_lower = query.to_lowercase();
+    let mut hits = 0;
+
+    for project in projects {
+        for session in &project.sessions {
+            let messages = match parse_session_deduped(&session.jsonl_path) {
+                Ok(m) => m,
+                Err(_) => continue,
+            };
+
+            for msg in &messages {
+                let text = match msg {
+                    types::RawMessage::User(u) => match &u.message.content {
+                        types::Content::Text(s) => s.clone(),
+                        types::Content::Blocks(_) => continue,
+                    },
+                    types::RawMessage::Assistant(a) => {
+                        let mut parts = Vec::new();
+                        for block in &a.message.content {
+                            if let types::ContentBlock::Text { text } = block {
+                                parts.push(text.as_str());
+                            }
+                        }
+                        parts.join(" ")
+                    }
+                    _ => continue,
+                };
+
+                if text.to_lowercase().contains(&query_lower) {
+                    let role = match msg {
+                        types::RawMessage::User(_) => "user",
+                        types::RawMessage::Assistant(_) => "assistant",
+                        _ => "other",
+                    };
+                    let preview = truncate(&text, 120);
+                    println!(
+                        "[{}] {} {} | {}",
+                        &project.original_path,
+                        &session.id[..8.min(session.id.len())],
+                        role,
+                        preview
+                    );
+                    hits += 1;
+                }
+            }
+        }
+    }
+
+    println!("\n{hits} matches for \"{query}\"");
+}
+
+/// Show aggregate statistics across all sessions.
+fn cmd_stats(projects: &[ClaudeProject]) {
+    let mut total_sessions = 0;
+    let mut total_messages = 0;
+    let mut total_tools = 0;
+    let mut total_subagents = 0;
+    let mut tool_counts: std::collections::HashMap<String, usize> = std::collections::HashMap::new();
+
+    for project in projects {
+        for session in &project.sessions {
+            total_sessions += 1;
+            if let Ok(parsed) = load_session(session) {
+                total_messages += parsed.messages.len();
+                total_tools += parsed.tool_exchanges.len();
+                total_subagents += parsed.subagents.len();
+
+                for ex in &parsed.tool_exchanges {
+                    let name = format!("{:?}", ex.call.name);
+                    *tool_counts.entry(name).or_default() += 1;
+                }
+            }
+        }
+    }
+
+    println!("--- Statistics ---");
+    println!("Projects:    {}", projects.len());
+    println!("Sessions:    {total_sessions}");
+    println!("Messages:    {total_messages}");
+    println!("Tool calls:  {total_tools}");
+    println!("Sub-agents:  {total_subagents}");
+
+    if !tool_counts.is_empty() {
+        println!("\nTool usage:");
+        let mut sorted: Vec<_> = tool_counts.into_iter().collect();
+        sorted.sort_by(|a, b| b.1.cmp(&a.1));
+        for (name, count) in sorted.iter().take(15) {
+            println!("  {name:20} {count}");
+        }
+    }
+}
+
+fn truncate(s: &str, max: usize) -> String {
+    let s = s.replace('\n', " ").replace('\r', "");
+    if s.len() <= max {
+        s
+    } else {
+        format!("{}...", &s[..max])
+    }
+}

crates/dirigent_anth/src/bin/anth_usage.rs

@@ -0,0 +1,192 @@
+use portable_pty::{CommandBuilder, NativePtySystem, PtySize, PtySystem};
+use std::io::{Read, Write};
+use std::path::PathBuf;
+use std::time::Duration;
+
+const ROWS: u16 = 80;
+const COLS: u16 = 120;
+
+struct Args {
+    debug: bool,
+    raw: bool,
+    no_trust: bool,
+    workdir: Option<PathBuf>,
+    use_cwd: bool,
+}
+
+fn parse_args() -> Args {
+    let mut args = Args {
+        debug: false,
+        raw: false,
+        no_trust: false,
+        workdir: None,
+        use_cwd: false,
+    };
+    let mut iter = std::env::args().skip(1);
+    while let Some(arg) = iter.next() {
+        match arg.as_str() {
+            "--debug" => args.debug = true,
+            "--raw" => args.raw = true,
+            "--no-trust" => args.no_trust = true,
+            "--cwd" => args.use_cwd = true,
+            "--workdir" => {
+                args.workdir = Some(PathBuf::from(
+                    iter.next().expect("--workdir requires a path argument"),
+                ));
+            }
+            other => {
+                eprintln!("Unknown argument: {other}");
+                eprintln!(
+                    "Usage: anth_usage [--debug] [--raw] [--no-trust] [--workdir <path>] [--cwd]"
+                );
+                std::process::exit(2);
+            }
+        }
+    }
+    args
+}
+
+fn resolve_workdir(args: &Args) -> PathBuf {
+    if let Some(ref dir) = args.workdir {
+        return dir.clone();
+    }
+    if args.use_cwd {
+        return std::env::current_dir().expect("failed to get current directory");
+    }
+
+    #[cfg(feature = "dirigent-paths")]
+    {
+        if let Ok(paths) = dirigent_config::DirigentPaths::resolve() {
+            let noproject = paths.noproject_home_dir();
+            if noproject.exists() {
+                return noproject;
+            }
+        }
+    }
+
+    dirs::home_dir().expect("failed to resolve home directory")
+}
+
+fn grab_screen(parser: &vt100::Parser) -> String {
+    let screen = parser.screen();
+    let mut output = String::new();
+    for line in screen.rows(0, COLS) {
+        output.push_str(&line);
+        output.push('\n');
+    }
+    output
+}
+
+macro_rules! debug {
+    ($args:expr, $($tt:tt)*) => {
+        if $args.debug {
+            eprintln!($($tt)*);
+        }
+    };
+}
+
+fn main() {
+    let args = parse_args();
+    let workdir = resolve_workdir(&args);
+
+    debug!(args, "Working directory: {}", workdir.display());
+
+    let pty_system = NativePtySystem::default();
+    let pair = pty_system
+        .openpty(PtySize {
+            rows: ROWS,
+            cols: COLS,
+            pixel_width: 0,
+            pixel_height: 0,
+        })
+        .expect("failed to open pty");
+
+    let mut cmd = CommandBuilder::new("claude");
+    cmd.cwd(&workdir);
+    let mut child = pair.slave.spawn_command(cmd).expect("failed to spawn claude");
+    drop(pair.slave);
+
+    let mut writer = pair.master.take_writer().expect("failed to get writer");
+    let reader = pair.master.try_clone_reader().expect("failed to get reader");
+
+    let (tx, rx) = std::sync::mpsc::channel();
+    std::thread::spawn(move || {
+        let mut reader = reader;
+        let mut buf = [0u8; 4096];
+        loop {
+            match reader.read(&mut buf) {
+                Ok(0) => break,
+                Ok(n) => {
+                    let _ = tx.send(buf[..n].to_vec());
+                }
+                Err(_) => break,
+            }
+        }
+    });
+
+    // Wait for claude to render
+    std::thread::sleep(Duration::from_secs(5));
+
+    debug!(
+        args,
+        "Child alive: {}",
+        matches!(child.try_wait(), Ok(None))
+    );
+
+    // Grab screen
+    let mut parser = vt100::Parser::new(ROWS, COLS, 0);
+    while let Ok(data) = rx.try_recv() {
+        parser.process(&data);
+    }
+    let output = grab_screen(&parser);
+    debug!(args, "=== SCREEN ===\n{output}=== END ===");
+
+    // Handle trust prompt
+    if output.contains("Yes, I trust this folder") {
+        if args.no_trust {
+            eprintln!("Folder is not trusted: {}", workdir.display());
+            eprintln!("Run claude in this folder manually to trust it, or omit --no-trust.");
+            let _ = child.kill();
+            std::process::exit(1);
+        }
+        debug!(args, "Sending enter for trust...");
+        writer.write_all(b"\r").expect("failed to confirm trust");
+
+        std::thread::sleep(Duration::from_secs(3));
+
+        while let Ok(data) = rx.try_recv() {
+            parser.process(&data);
+        }
+        debug!(
+            args,
+            "=== AFTER TRUST ===\n{}=== END ===",
+            grab_screen(&parser)
+        );
+    }
+
+    // Send /usage
+    debug!(args, "Sending /usage...");
+    writer
+        .write_all(b"/usage\r")
+        .expect("failed to send /usage");
+
+    std::thread::sleep(Duration::from_secs(3));
+
+    while let Ok(data) = rx.try_recv() {
+        parser.process(&data);
+    }
+    let raw_output = grab_screen(&parser);
+
+    let processed = dirigent_anth::anth_usage::process_usage_screen(&raw_output);
+
+    if args.raw {
+        println!("{}", processed.raw_screen);
+    } else {
+        println!(
+            "{}",
+            serde_json::to_string_pretty(&processed.data).expect("failed to serialize usage data")
+        );
+    }
+
+    let _ = child.kill();
+}

crates/dirigent_anth/src/claude_grab.rs

@@ -0,0 +1,157 @@
+use portable_pty::{Child, CommandBuilder, NativePtySystem, PtySize, PtySystem};
+use std::io::{Read, Write};
+use std::sync::mpsc::{self, Receiver};
+use std::time::Duration;
+use vt100::Parser;
+
+const DEFAULT_ROWS: u16 = 80;
+const DEFAULT_COLS: u16 = 120;
+
+pub struct PtySession {
+    parser: Parser,
+    writer: Option<Box<dyn Write + Send>>,
+    rx: Receiver<Vec<u8>>,
+    cols: u16,
+    #[allow(dead_code)]
+    child: Box<dyn Child + Send + Sync>,
+}
+
+impl PtySession {
+    pub fn spawn_claude(args: &[&str]) -> Self {
+        Self::spawn_claude_with_size(args, DEFAULT_ROWS, DEFAULT_COLS)
+    }
+
+    pub fn spawn_claude_with_size(args: &[&str], rows: u16, cols: u16) -> Self {
+        let pty_system = NativePtySystem::default();
+
+        let pair = pty_system
+            .openpty(PtySize {
+                rows,
+                cols,
+                pixel_width: 0,
+                pixel_height: 0,
+            })
+            .expect("failed to open pty");
+
+        let mut cmd = CommandBuilder::new("claude");
+        for arg in args {
+            cmd.arg(*arg);
+        }
+        if let Some(home) = dirs::home_dir() {
+            cmd.cwd(home);
+        }
+        let child = pair
+            .slave
+            .spawn_command(cmd)
+            .expect("failed to spawn claude");
+
+        drop(pair.slave);
+
+        let writer = pair.master.take_writer().expect("failed to get writer");
+        let reader = pair
+            .master
+            .try_clone_reader()
+            .expect("failed to get reader");
+
+        let (tx, rx) = mpsc::channel::<Vec<u8>>();
+        std::thread::spawn(move || {
+            let mut reader = reader;
+            let mut chunk = [0u8; 4096];
+            loop {
+                match reader.read(&mut chunk) {
+                    Ok(0) => break,
+                    Ok(n) => {
+                        if tx.send(chunk[..n].to_vec()).is_err() {
+                            break;
+                        }
+                    }
+                    Err(_) => break,
+                }
+            }
+        });
+
+        Self {
+            parser: Parser::new(rows, cols, 0),
+            writer: Some(writer),
+            rx,
+            cols,
+            child,
+        }
+    }
+
+    pub fn grab_screen(&mut self) -> String {
+        while let Ok(data) = self.rx.try_recv() {
+            self.parser.process(&data);
+        }
+        let deadline = std::time::Instant::now() + Duration::from_millis(200);
+        while std::time::Instant::now() < deadline {
+            match self.rx.recv_timeout(Duration::from_millis(50)) {
+                Ok(data) => self.parser.process(&data),
+                Err(_) => {}
+            }
+        }
+
+        let screen = self.parser.screen();
+        let mut output = String::new();
+        for line in screen.rows(0, self.cols) {
+            output.push_str(&line);
+            output.push('\n');
+        }
+        output
+    }
+
+    pub fn wait_for(&mut self, needle: &str, timeout: Duration) -> bool {
+        self.wait_for_any(&[needle], timeout)
+    }
+
+    pub fn wait_for_any(&mut self, needles: &[&str], timeout: Duration) -> bool {
+        let deadline = std::time::Instant::now() + timeout;
+        while std::time::Instant::now() < deadline {
+            match self.rx.recv_timeout(Duration::from_millis(100)) {
+                Ok(data) => self.parser.process(&data),
+                Err(_) => {}
+            }
+            let screen = self.parser.screen();
+            let mut content = String::new();
+            for line in screen.rows(0, self.cols) {
+                content.push_str(&line);
+                content.push('\n');
+            }
+            for needle in needles {
+                if content.contains(needle) {
+                    return true;
+                }
+            }
+        }
+        false
+    }
+
+    pub fn is_alive(&mut self) -> bool {
+        matches!(self.child.try_wait(), Ok(None))
+    }
+
+    pub fn send(&mut self, input: &[u8]) {
+        self.writer.as_mut().expect("writer gone").write_all(input).expect("failed to write to pty");
+    }
+
+    pub fn try_send(&mut self, input: &[u8]) -> std::io::Result<()> {
+        match self.writer.as_mut() {
+            Some(w) => w.write_all(input),
+            None => Err(std::io::Error::new(std::io::ErrorKind::BrokenPipe, "writer gone")),
+        }
+    }
+
+    pub fn try_send_line(&mut self, text: &str) -> std::io::Result<()> {
+        self.try_send(text.as_bytes())?;
+        self.try_send(b"\r")
+    }
+
+    pub fn send_enter(&mut self) {
+        self.send(b"\r");
+    }
+
+    pub fn send_line(&mut self, text: &str) {
+        self.send(text.as_bytes());
+        self.send_enter();
+    }
+}

crates/dirigent_anth/src/correlation.rs

@@ -0,0 +1,107 @@
+//! Tool call correlation — matches assistant ToolUse blocks with their
+//! corresponding user ToolResult blocks by ID across a message sequence.
+
+use std::collections::HashMap;
+
+use crate::types::{
+    Content, ContentBlock, RawAssistantMessage, RawMessage, RawUserMessage, ToolCall,
+    ToolExchange, ToolName, ToolResultData,
+};
+
+/// Extract tool calls from an assistant message's content blocks.
+fn extract_tool_calls(msg: &RawAssistantMessage) -> Vec<ToolCall> {
+    let source_uuid = msg.uuid.clone().unwrap_or_default();
+    msg.message
+        .content
+        .iter()
+        .filter_map(|block| {
+            if let ContentBlock::ToolUse { id, name, input, .. } = block {
+                Some(ToolCall {
+                    id: id.clone(),
+                    name: ToolName::from(name.clone()),
+                    input: input.clone(),
+                    source_message_uuid: source_uuid.clone(),
+                })
+            } else {
+                None
+            }
+        })
+        .collect()
+}
+
+/// Extract tool results from a user message's content blocks.
+fn extract_tool_results(msg: &RawUserMessage) -> Vec<ToolResultData> {
+    let source_uuid = msg.uuid.clone().unwrap_or_default();
+    match &msg.message.content {
+        Content::Blocks(blocks) => blocks
+            .iter()
+            .filter_map(|block| {
+                if let ContentBlock::ToolResult { tool_use_id, content, is_error } = block {
+                    // Extract text content from the tool result
+                    let text_content = content.as_ref().and_then(|c| match c {
+                        Content::Text(s) => Some(s.clone()),
+                        Content::Blocks(bs) => {
+                            // Concatenate text blocks
+                            let texts: Vec<&str> = bs
+                                .iter()
+                                .filter_map(|b| {
+                                    if let ContentBlock::Text { text } = b {
+                                        Some(text.as_str())
+                                    } else {
+                                        None
+                                    }
+                                })
+                                .collect();
+                            if texts.is_empty() { None } else { Some(texts.join("\n")) }
+                        }
+                    });
+                    Some(ToolResultData {
+                        tool_use_id: tool_use_id.clone(),
+                        content: text_content,
+                        is_error: *is_error,
+                        source_message_uuid: source_uuid.clone(),
+                    })
+                } else {
+                    None
+                }
+            })
+            .collect(),
+        Content::Text(_) => Vec::new(),
+    }
+}
+
+/// Correlate tool calls with their results across a message sequence.
+///
+/// Iterates messages in order, collecting ToolUse blocks from assistant
+/// messages and matching them by ID to ToolResult blocks in subsequent user
+/// messages. Any tool calls that never received a result are emitted with
+/// `result: None`.
+pub fn correlate_tools(messages: &[RawMessage]) -> Vec<ToolExchange> {
+    let mut pending: HashMap<String, ToolCall> = HashMap::new();
+    let mut exchanges: Vec<ToolExchange> = Vec::new();
+
+    for msg in messages {
+        match msg {
+            RawMessage::Assistant(asst) => {
+                for call in extract_tool_calls(asst) {
+                    pending.insert(call.id.clone(), call);
+                }
+            }
+            RawMessage::User(user) => {
+                for result in extract_tool_results(user) {
+                    if let Some(call) = pending.remove(&result.tool_use_id) {
+                        exchanges.push(ToolExchange { call, result: Some(result) });
+                    }
+                }
+            }
+            _ => {}
+        }
+    }
+
+    // Emit unmatched calls (no result found)
+    for (_id, call) in pending {
+        exchanges.push(ToolExchange { call, result: None });
+    }
+
+    exchanges
+}

crates/dirigent_anth/src/dedup.rs

@@ -0,0 +1,116 @@
+//! Streaming deduplication for assistant messages.
+
+use crate::types::{RawAssistantMessage, RawMessage};
+
+/// Deduplicate streamed assistant messages.
+///
+/// Claude Code writes multiple JSONL lines for the same assistant message
+/// as it streams. Each shares the same `uuid` with progressively more
+/// content blocks. We keep only the last entry per uuid.
+///
+/// Non-assistant messages pass through unchanged.
+pub fn dedup_messages(messages: Vec<RawMessage>) -> Vec<RawMessage> {
+    let mut result: Vec<RawMessage> = Vec::new();
+    let mut buffered_assistant: Option<RawAssistantMessage> = None;
+
+    for msg in messages {
+        match msg {
+            RawMessage::Assistant(ref asst) => {
+                let current_uuid = asst.uuid.as_deref();
+
+                if let Some(ref buffered) = buffered_assistant {
+                    let buffered_uuid = buffered.uuid.as_deref();
+                    if current_uuid == buffered_uuid {
+                        // Same uuid — replace buffer with newer (more complete) version
+                        buffered_assistant = Some(asst.clone());
+                    } else {
+                        // Different uuid — flush old buffer, start new
+                        result.push(RawMessage::Assistant(buffered.clone()));
+                        buffered_assistant = Some(asst.clone());
+                    }
+                } else {
+                    // No buffer yet — start buffering
+                    buffered_assistant = Some(asst.clone());
+                }
+            }
+            _ => {
+                // Non-assistant: flush any buffered assistant first, then push this
+                if let Some(buffered) = buffered_assistant.take() {
+                    result.push(RawMessage::Assistant(buffered));
+                }
+                result.push(msg);
+            }
+        }
+    }
+
+    // Flush remaining buffer
+    if let Some(buffered) = buffered_assistant {
+        result.push(RawMessage::Assistant(buffered));
+    }
+
+    result
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::types::{AssistantInner, ContentBlock};
+
+    fn make_assistant(uuid: &str, stop_reason: Option<&str>, text: &str) -> RawMessage {
+        RawMessage::Assistant(RawAssistantMessage {
+            uuid: Some(uuid.to_string()),
+            parent_uuid: None,
+            timestamp: None,
+            session_id: None,
+            cwd: None,
+            version: None,
+            git_branch: None,
+            is_sidechain: false,
+            request_id: None,
+            message: AssistantInner {
+                model: None,
+                id: None,
+                message_type: None,
+                role: None,
+                content: vec![ContentBlock::Text { text: text.to_string() }],
+                stop_reason: stop_reason.map(str::to_string),
+                stop_sequence: None,
+                usage: None,
+            },
+        })
+    }
+
+    #[test]
+    fn dedup_single_streamed_message() {
+        let msgs = vec![
+            make_assistant("a-1", None, "Part 1"),
+            make_assistant("a-1", None, "Part 1 more"),
+            make_assistant("a-1", Some("end_turn"), "Part 1 final"),
+        ];
+        let deduped = dedup_messages(msgs);
+        assert_eq!(deduped.len(), 1);
+        if let RawMessage::Assistant(a) = &deduped[0] {
+            assert_eq!(a.message.stop_reason.as_deref(), Some("end_turn"));
+            match &a.message.content[0] {
+                ContentBlock::Text { text } => assert_eq!(text, "Part 1 final"),
+                _ => panic!("Expected text block"),
+            }
+        }
+    }
+
+    #[test]
+    fn dedup_two_distinct_assistants() {
+        let msgs = vec![
+            make_assistant("a-1", Some("end_turn"), "First"),
+            make_assistant("a-2", Some("end_turn"), "Second"),
+        ];
+        let deduped = dedup_messages(msgs);
+        assert_eq!(deduped.len(), 2);
+    }
+
+    #[test]
+    fn dedup_empty_input() {
+        let deduped = dedup_messages(vec![]);
+        assert!(deduped.is_empty());
+    }
+}

crates/dirigent_anth/src/discovery.rs

@@ -0,0 +1,342 @@
+use std::collections::HashMap;
+use camino::{Utf8Path, Utf8PathBuf};
+use crate::types::*;
+use crate::error::{AntError, Result};
+
+/// Discover the Claude Code home directory (~/.claude/).
+pub fn discover_claude_home() -> Result<Utf8PathBuf> {
+    let home = dirs::home_dir().ok_or(AntError::HomeNotFound)?;
+    let claude_dir = home.join(".claude");
+    if !claude_dir.exists() {
+        return Err(AntError::HomeNotFound);
+    }
+    Utf8PathBuf::try_from(claude_dir.to_path_buf())
+        .map_err(|e| AntError::InvalidPath(e.to_string()))
+}
+
+/// Normalise a native path to forward slashes for consistent storage.
+fn normalize_to_forward_slashes(path: &str) -> String {
+    path.replace('\\', "/")
+}
+
+/// Resolve the original filesystem path for a Claude project directory.
+///
+/// Priority:
+/// 1. `projectPath` from `sessions-index.json` (authoritative, cheap)
+/// 2. `cwd` from the first user message in any session JSONL (authoritative, costs one file parse)
+/// 3. `decode_project_path` (lossy fallback for empty project directories)
+pub fn resolve_original_path(dir_name: &str, sessions: &[SessionRef]) -> String {
+    // 1. Try sessions-index.json projectPath
+    for session in sessions {
+        if let Some(ref idx) = session.index_entry {
+            if let Some(ref path) = idx.project_path {
+                if !path.is_empty() {
+                    return normalize_to_forward_slashes(path);
+                }
+            }
+        }
+    }
+
+    // 2. Try cwd from first user message in any session
+    for session in sessions {
+        if let Ok(msgs) = crate::parser::parse_session(&session.jsonl_path) {
+            for msg in &msgs {
+                if let crate::types::RawMessage::User(user) = msg {
+                    if let Some(ref cwd) = user.cwd {
+                        if !cwd.is_empty() {
+                            return normalize_to_forward_slashes(cwd);
+                        }
+                    }
+                }
+            }
+        }
+    }
+
+    // 3. Lossy fallback
+    decode_project_path(dir_name)
+}
+
+/// Discover all Claude Code project directories under the given home.
+pub fn discover_projects(home: &Utf8Path) -> Result<Vec<ClaudeProject>> {
+    let projects_dir = home.join("projects");
+    if !projects_dir.as_std_path().exists() {
+        return Ok(Vec::new());
+    }
+
+    let mut projects = Vec::new();
+    for entry in std::fs::read_dir(projects_dir.as_std_path())? {
+        let entry = entry?;
+        let path = entry.path();
+        if !path.is_dir() {
+            continue;
+        }
+        let dir_name = match path.file_name().and_then(|n| n.to_str()) {
+            Some(name) => name.to_string(),
+            None => continue,
+        };
+
+        let utf8_path = match Utf8PathBuf::try_from(path.clone()) {
+            Ok(p) => p,
+            Err(_) => continue,
+        };
+
+        let sessions = discover_sessions(&utf8_path)?;
+        let original_path = resolve_original_path(&dir_name, &sessions);
+
+        projects.push(ClaudeProject {
+            path: utf8_path,
+            original_path,
+            sessions,
+        });
+    }
+
+    Ok(projects)
+}
+
+/// Decode an encoded project folder name back to the original path (lossy).
+///
+/// **Warning**: Claude Code's encoding replaces `\`, `/`, AND `_` all with
+/// `-`, making this decoding ambiguous. For example, `G--dev-projects-adk-rust`
+/// could be `G:/dev/projects/adk-rust` or `G:/dev/projects/adk/rust`. Prefer
+/// [`resolve_original_path`] which reads ground truth from `sessions-index.json`
+/// or session JSONL files. This function is a last-resort fallback for empty
+/// project directories with no sessions or index.
+pub fn decode_project_path(encoded: &str) -> String {
+    // Split on "--" to recover path segments separated by the original separators.
+    let parts: Vec<&str> = encoded.split("--").collect();
+
+    if parts.is_empty() {
+        return encoded.to_string();
+    }
+
+    let mut result = String::new();
+
+    let first = parts[0];
+
+    if first.len() == 1 && first.chars().next().map_or(false, |c| c.is_ascii_uppercase()) {
+        // Windows drive letter: "G" → "G:"
+        result.push_str(first);
+        result.push(':');
+    } else if first.starts_with('-') || first.is_empty() {
+        // Unix-style absolute path: the original path started with "/".
+        // The first segment has a leading "-" that encoded the root separator.
+        // Strip that leading "-" to recover the first directory component.
+        let component = first.trim_start_matches('-');
+        result.push('/');
+        if !component.is_empty() {
+            // Single dashes within the component are path separators.
+            result.push_str(&component.replace('-', "/"));
+        }
+    } else {
+        result.push_str(first);
+    }
+
+    // Remaining "--"-separated parts are additional path components.
+    // Within each part, single "-" represent path separators.
+    for part in &parts[1..] {
+        result.push('/');
+        result.push_str(&part.replace('-', "/"));
+    }
+
+    result
+}
+
+/// Discover all session JSONL files in a project directory.
+pub fn discover_sessions(project_dir: &Utf8Path) -> Result<Vec<SessionRef>> {
+    let index = load_session_index(project_dir);
+    let mut sessions = Vec::new();
+
+    for entry in std::fs::read_dir(project_dir.as_std_path())? {
+        let entry = entry?;
+        let path = entry.path();
+
+        // Only .jsonl files
+        let extension = path.extension().and_then(|e| e.to_str());
+        if extension != Some("jsonl") {
+            continue;
+        }
+
+        let stem = match path.file_stem().and_then(|s| s.to_str()) {
+            Some(s) => s.to_string(),
+            None => continue,
+        };
+
+        let utf8_path = match Utf8PathBuf::try_from(path) {
+            Ok(p) => p,
+            Err(_) => continue,
+        };
+
+        // Check for artifacts directory (same name as the session stem).
+        let artifacts_dir = {
+            let dir = project_dir.join(&stem);
+            if dir.as_std_path().is_dir() {
+                Some(dir)
+            } else {
+                None
+            }
+        };
+
+        let index_entry = index.as_ref().and_then(|idx| idx.get(&stem).cloned());
+
+        sessions.push(SessionRef {
+            id: stem,
+            jsonl_path: utf8_path,
+            artifacts_dir,
+            index_entry,
+        });
+    }
+
+    Ok(sessions)
+}
+
+/// Load `sessions-index.json` if it exists in the given project directory.
+fn load_session_index(project_dir: &Utf8Path) -> Option<HashMap<String, SessionIndexEntry>> {
+    let index_path = project_dir.join("sessions-index.json");
+    if !index_path.as_std_path().exists() {
+        return None;
+    }
+
+    let content = std::fs::read_to_string(index_path.as_std_path()).ok()?;
+    serde_json::from_str::<HashMap<String, SessionIndexEntry>>(&content).ok()
+}
+
+// ---------------------------------------------------------------------------
+// Tests
+// ---------------------------------------------------------------------------
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use tempfile::TempDir;
+
+    #[test]
+    fn decode_project_path_windows() {
+        assert_eq!(
+            decode_project_path("G--dev-projects-dirigent"),
+            "G:/dev/projects/dirigent"
+        );
+    }
+
+    #[test]
+    fn decode_project_path_windows_users() {
+        assert_eq!(
+            decode_project_path("C--Users-g4b-tmp"),
+            "C:/Users/g4b/tmp"
+        );
+    }
+
+    #[test]
+    fn decode_project_path_unix() {
+        assert_eq!(
+            decode_project_path("-home-user-projects-foo"),
+            "/home/user/projects/foo"
+        );
+    }
+
+    #[test]
+    fn discover_sessions_in_temp_dir() {
+        let tmp = TempDir::new().unwrap();
+        let project_dir = Utf8Path::from_path(tmp.path()).unwrap();
+
+        // Create fake session files.
+        std::fs::write(project_dir.join("abc-def-123.jsonl").as_std_path(), "{}\n").unwrap();
+        std::fs::write(project_dir.join("xyz-456-789.jsonl").as_std_path(), "{}\n").unwrap();
+        // Create an artifacts directory for one session.
+        std::fs::create_dir(project_dir.join("abc-def-123").as_std_path()).unwrap();
+
+        let sessions = discover_sessions(project_dir).unwrap();
+        assert_eq!(sessions.len(), 2);
+
+        let with_artifacts = sessions.iter().find(|s| s.id == "abc-def-123").unwrap();
+        assert!(with_artifacts.artifacts_dir.is_some());
+
+        let without_artifacts = sessions.iter().find(|s| s.id == "xyz-456-789").unwrap();
+        assert!(without_artifacts.artifacts_dir.is_none());
+    }
+
+    #[test]
+    fn discover_sessions_ignores_non_jsonl() {
+        let tmp = TempDir::new().unwrap();
+        let project_dir = Utf8Path::from_path(tmp.path()).unwrap();
+
+        std::fs::write(project_dir.join("session.jsonl").as_std_path(), "{}\n").unwrap();
+        std::fs::write(
+            project_dir.join("sessions-index.json").as_std_path(),
+            "{}",
+        )
+        .unwrap();
+        std::fs::create_dir(project_dir.join("some-dir").as_std_path()).unwrap();
+
+        let sessions = discover_sessions(project_dir).unwrap();
+        assert_eq!(sessions.len(), 1);
+        assert_eq!(sessions[0].id, "session");
+    }
+
+    #[test]
+    fn discover_sessions_loads_index_entry() {
+        let tmp = TempDir::new().unwrap();
+        let project_dir = Utf8Path::from_path(tmp.path()).unwrap();
+
+        std::fs::write(project_dir.join("abc-123.jsonl").as_std_path(), "{}\n").unwrap();
+
+        let index_json = r#"{
+            "abc-123": {
+                "sessionId": "abc-123",
+                "firstPrompt": "Hello",
+                "summary": "A test session",
+                "messageCount": 5
+            }
+        }"#;
+        std::fs::write(
+            project_dir.join("sessions-index.json").as_std_path(),
+            index_json,
+        )
+        .unwrap();
+
+        let sessions = discover_sessions(project_dir).unwrap();
+        assert_eq!(sessions.len(), 1);
+
+        let entry = sessions[0].index_entry.as_ref().unwrap();
+        assert_eq!(entry.session_id.as_deref(), Some("abc-123"));
+        assert_eq!(entry.first_prompt.as_deref(), Some("Hello"));
+        assert_eq!(entry.message_count, Some(5));
+    }
+
+    #[test]
+    fn resolve_original_path_prefers_index_project_path() {
+        let sessions = vec![SessionRef {
+            id: "test-session".to_string(),
+            jsonl_path: Utf8PathBuf::from("/tmp/fake.jsonl"),
+            artifacts_dir: None,
+            index_entry: Some(SessionIndexEntry {
+                session_id: Some("test-session".to_string()),
+                first_prompt: None,
+                summary: None,
+                message_count: None,
+                created: None,
+                modified: None,
+                git_branch: None,
+                project_path: Some(r"G:\dev\projects\bevy_sprite3d".to_string()),
+            }),
+        }];
+        let result = resolve_original_path("G--dev-projects-bevy-sprite3d", &sessions);
+        assert_eq!(result, "G:/dev/projects/bevy_sprite3d");
+    }
+
+    #[test]
+    fn resolve_original_path_falls_back_to_decode() {
+        let sessions: Vec<SessionRef> = vec![];
+        let result = resolve_original_path("G--dev-projects-dirigent", &sessions);
+        assert_eq!(result, "G:/dev/projects/dirigent");
+    }
+
+    #[test]
+    fn discover_projects_empty_when_no_projects_dir() {
+        let tmp = TempDir::new().unwrap();
+        let home_dir = Utf8Path::from_path(tmp.path()).unwrap();
+
+        // No "projects" subdirectory — should return empty vec, not an error.
+        let projects = discover_projects(home_dir).unwrap();
+        assert!(projects.is_empty());
+    }
+}

crates/dirigent_anth/src/error.rs

@@ -0,0 +1,19 @@
+#[derive(Debug, thiserror::Error)]
+pub enum AntError {
+    #[error("I/O error: {0}")]
+    Io(#[from] std::io::Error),
+
+    #[error("JSON parse error at line {line}: {source}")]
+    JsonParse {
+        line: usize,
+        source: serde_json::Error,
+    },
+
+    #[error("Claude home directory not found")]
+    HomeNotFound,
+
+    #[error("Invalid path: {0}")]
+    InvalidPath(String),
+}
+
+pub type Result<T> = std::result::Result<T, AntError>;

crates/dirigent_anth/src/lib.rs

@@ -0,0 +1,52 @@
+//! dirigent_anth — Claude Code Session Parser & Toolkit
+//!
+//! Reads Claude Code's local JSONL session storage and produces typed,
+//! deduplicated, correlated Rust data structures.
+//!
+//! # Design
+//!
+//! See `docs/superpowers/plans/2026-03-23-dirigent-ant-design.md`
+
+pub mod claude_grab;
+pub mod anth_usage;
+pub mod correlation;
+pub mod dedup;
+pub mod discovery;
+pub mod error;
+pub mod noise;
+pub mod parser;
+pub mod subagent;
+pub mod tree;
+pub mod types;
+pub mod util;
+
+/// Load and fully parse a session: dedup, correlate, tree, subagents.
+pub fn load_session(session_ref: &types::SessionRef) -> error::Result<types::ParsedSession> {
+    let messages = parser::parse_session_deduped(&session_ref.jsonl_path)?;
+    let tree = tree::ConversationTree::build(&messages);
+    let tool_exchanges = correlation::correlate_tools(&messages);
+    let mut subagents = if let Some(ref dir) = session_ref.artifacts_dir {
+        subagent::load_subagents(dir)?
+    } else {
+        Vec::new()
+    };
+    subagent::link_subagents_to_calls(&mut subagents, &tool_exchanges);
+
+    Ok(types::ParsedSession {
+        messages,
+        tree,
+        tool_exchanges,
+        subagents,
+    })
+}
+
+pub use correlation::correlate_tools;
+pub use dedup::dedup_messages;
+pub use discovery::{decode_project_path, discover_claude_home, discover_projects, discover_sessions, resolve_original_path};
+pub use error::{AntError, Result};
+pub use noise::{classify_noise, NoiseKind};
+pub use parser::{parse_line, parse_session, parse_session_deduped};
+pub use subagent::{link_subagents_to_calls, load_subagents};
+pub use tree::{message_parent_uuid, message_uuid, ConversationNode, ConversationTree};
+pub use types::*;
+pub use util::parse_timestamp;

crates/dirigent_anth/src/noise.rs

@@ -0,0 +1,72 @@
+use crate::types::*;
+
+/// Classification of noise patterns in Claude Code JSONL.
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub enum NoiseKind {
+    Meta,
+    Warmup,
+    Interrupted,
+    Continuation,
+    ApiError,
+    SystemCaveat,
+    QueueOp,
+}
+
+/// Classify a message as noise, if applicable.
+/// Returns None for normal messages.
+pub fn classify_noise(message: &RawMessage) -> Option<NoiseKind> {
+    match message {
+        RawMessage::QueueOperation(_) => Some(NoiseKind::QueueOp),
+        RawMessage::User(user) => {
+            if user.is_meta.unwrap_or(false) {
+                return Some(NoiseKind::Meta);
+            }
+            if let Some(text) = extract_user_text(user) {
+                if text == "Warmup" {
+                    return Some(NoiseKind::Warmup);
+                }
+                if text.starts_with("[Request interrupted") {
+                    return Some(NoiseKind::Interrupted);
+                }
+                if text.starts_with("This session is being continued") {
+                    return Some(NoiseKind::Continuation);
+                }
+                if text.starts_with("API Error") {
+                    return Some(NoiseKind::ApiError);
+                }
+                if text.starts_with("Caveat: The messages below") {
+                    return Some(NoiseKind::SystemCaveat);
+                }
+            }
+            None
+        }
+        _ => None,
+    }
+}
+
+/// Extract plain text from a user message's content.
+fn extract_user_text(user: &RawUserMessage) -> Option<&str> {
+    match &user.message.content {
+        Content::Text(s) => Some(s.as_str()),
+        Content::Blocks(_) => None, // tool_result blocks, not plain text
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn normal_assistant_is_not_noise() {
+        let json = r#"{"type":"assistant","uuid":"x","timestamp":"2026-01-01T00:00:00Z","sessionId":"s","message":{"id":"m","role":"assistant","content":[{"type":"text","text":"Hello"}],"stop_reason":"end_turn"}}"#;
+        let msg: RawMessage = serde_json::from_str(json).unwrap();
+        assert_eq!(classify_noise(&msg), None);
+    }
+
+    #[test]
+    fn queue_op_is_noise() {
+        let json = r#"{"type":"queue-operation","operation":"enqueue","timestamp":"2026-01-01T00:00:00Z","sessionId":"s"}"#;
+        let msg: RawMessage = serde_json::from_str(json).unwrap();
+        assert_eq!(classify_noise(&msg), Some(NoiseKind::QueueOp));
+    }
+}

crates/dirigent_anth/src/parser.rs

@@ -0,0 +1,50 @@
+//! JSONL line parser for Claude Code session files.
+
+use std::io::BufRead;
+
+use camino::Utf8Path;
+
+use crate::error::Result;
+use crate::types::RawMessage;
+
+/// Parse a single JSONL line into a RawMessage.
+/// Returns None for lines that cannot be parsed (logged via tracing).
+pub fn parse_line(line: &str, line_number: usize) -> Option<RawMessage> {
+    match serde_json::from_str::<RawMessage>(line) {
+        Ok(msg) => Some(msg),
+        Err(e) => {
+            tracing::warn!(line = line_number, error = %e, "Skipping unparseable JSONL line");
+            None
+        }
+    }
+}
+
+/// Parse all messages from a JSONL file.
+/// Skips unparseable lines (lenient). Returns I/O errors.
+pub fn parse_session(path: &Utf8Path) -> Result<Vec<RawMessage>> {
+    let file = std::fs::File::open(path.as_std_path())?;
+    let reader = std::io::BufReader::new(file);
+    let mut messages = Vec::new();
+
+    for (i, line) in reader.lines().enumerate() {
+        let line = line?;
+        if line.trim().is_empty() {
+            continue;
+        }
+        if let Some(msg) = parse_line(&line, i + 1) {
+            messages.push(msg);
+        }
+    }
+
+    Ok(messages)
+}
+
+/// Parse a session JSONL file with streaming deduplication applied.
+///
+/// Claude Code writes multiple JSONL lines for the same assistant message as
+/// it streams. This function collapses those into a single final version per
+/// uuid. See [`crate::dedup::dedup_messages`] for details.
+pub fn parse_session_deduped(path: &Utf8Path) -> Result<Vec<RawMessage>> {
+    let messages = parse_session(path)?;
+    Ok(crate::dedup::dedup_messages(messages))
+}

crates/dirigent_anth/src/subagent.rs

@@ -0,0 +1,215 @@
+//! Sub-agent session loading.
+//!
+//! Claude Code spawns sub-agents for Agent tool calls and stores their
+//! conversations under `<session-artifacts-dir>/subagents/`. Each sub-agent
+//! has a JSONL file and an optional `.meta.json` with metadata such as the
+//! agent type.
+
+use camino::Utf8Path;
+
+use crate::error::Result;
+use crate::parser::parse_session;
+use crate::types::{SubAgentMeta, SubAgentSession, ToolExchange};
+
+/// Load all sub-agent sessions from a session's artifacts directory.
+///
+/// Expects files at: `<session_artifacts_dir>/subagents/agent-<id>.jsonl`
+/// with optional companion: `<session_artifacts_dir>/subagents/agent-<id>.meta.json`
+///
+/// Returns an empty `Vec` if the `subagents/` subdirectory does not exist.
+pub fn load_subagents(session_artifacts_dir: &Utf8Path) -> Result<Vec<SubAgentSession>> {
+    let subagents_dir = session_artifacts_dir.join("subagents");
+    if !subagents_dir.as_std_path().exists() {
+        return Ok(Vec::new());
+    }
+
+    let mut subagents = Vec::new();
+
+    for entry in std::fs::read_dir(subagents_dir.as_std_path())? {
+        let entry = entry?;
+        let path = entry.path();
+
+        // Only process agent-*.jsonl files
+        let file_name = match path.file_name().and_then(|n| n.to_str()) {
+            Some(name) => name.to_string(),
+            None => continue,
+        };
+
+        if !file_name.starts_with("agent-") || !file_name.ends_with(".jsonl") {
+            continue;
+        }
+
+        // Extract agent ID: "agent-abc123.jsonl" → "abc123"
+        let agent_id = file_name
+            .strip_prefix("agent-")
+            .and_then(|s| s.strip_suffix(".jsonl"))
+            .unwrap_or(&file_name)
+            .to_string();
+
+        let jsonl_path = match camino::Utf8PathBuf::try_from(path.clone()) {
+            Ok(p) => p,
+            Err(_) => continue,
+        };
+
+        // Parse the JSONL session
+        let messages = parse_session(&jsonl_path)?;
+
+        // Try to load companion metadata file
+        let meta_path = path.with_file_name(format!("agent-{}.meta.json", agent_id));
+        let meta = if meta_path.exists() {
+            let content = std::fs::read_to_string(&meta_path)?;
+            serde_json::from_str::<SubAgentMeta>(&content)
+                .unwrap_or(SubAgentMeta { agent_type: None })
+        } else {
+            SubAgentMeta { agent_type: None }
+        };
+
+        subagents.push(SubAgentSession {
+            agent_id,
+            meta,
+            messages,
+            parent_tool_call_id: None,
+        });
+    }
+
+    Ok(subagents)
+}
+
+/// Try to link sub-agent sessions to their parent Agent tool calls.
+///
+/// For each Agent tool call in `tool_exchanges`, parses the tool result text
+/// for `agentId: <id>` and matches it against sub-agent sessions. On match,
+/// sets `SubAgentSession.parent_tool_call_id` to the tool call's ID.
+///
+/// This is best-effort: if the agentId text format changes or a result is
+/// missing, the sub-agent is still usable but without tool_use linkage.
+pub fn link_subagents_to_calls(
+    subagents: &mut [SubAgentSession],
+    tool_exchanges: &[ToolExchange],
+) {
+    use regex::Regex;
+
+    if subagents.is_empty() || tool_exchanges.is_empty() {
+        return;
+    }
+
+    // Compile once, match many
+    let re = Regex::new(r"agentId:\s*(\S+)").expect("valid regex");
+
+    for exchange in tool_exchanges {
+        // Only look at Agent tool calls
+        if exchange.call.name != crate::types::ToolName::Agent {
+            continue;
+        }
+
+        // Extract agentId from the tool result text
+        let agent_id = exchange
+            .result
+            .as_ref()
+            .and_then(|r| r.content.as_deref())
+            .and_then(|text| re.captures(text))
+            .and_then(|caps| caps.get(1))
+            .map(|m| m.as_str());
+
+        let agent_id = match agent_id {
+            Some(id) => id,
+            None => continue,
+        };
+
+        // Find matching sub-agent and set the linkage
+        if let Some(subagent) = subagents.iter_mut().find(|s| s.agent_id == agent_id) {
+            subagent.parent_tool_call_id = Some(exchange.call.id.clone());
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::types::{ToolCall, ToolName, ToolResultData};
+
+    #[test]
+    fn test_link_subagents_to_calls_matches_agent_id() {
+        let mut subagents = vec![
+            SubAgentSession {
+                agent_id: "abc123def".to_string(),
+                meta: SubAgentMeta { agent_type: Some("Explore".to_string()) },
+                messages: vec![],
+                parent_tool_call_id: None,
+            },
+            SubAgentSession {
+                agent_id: "xyz789".to_string(),
+                meta: SubAgentMeta { agent_type: None },
+                messages: vec![],
+                parent_tool_call_id: None,
+            },
+        ];
+
+        let exchanges = vec![
+            ToolExchange {
+                call: ToolCall {
+                    id: "toolu_01ABC".to_string(),
+                    name: ToolName::Agent,
+                    input: serde_json::json!({"description": "test"}),
+                    source_message_uuid: "msg-1".to_string(),
+                },
+                result: Some(ToolResultData {
+                    tool_use_id: "toolu_01ABC".to_string(),
+                    content: Some("agentId: abc123def (use SendMessage with to: 'abc123def' to continue)\n<usage>total_tokens: 1000</usage>".to_string()),
+                    is_error: false,
+                    source_message_uuid: "msg-2".to_string(),
+                }),
+            },
+            ToolExchange {
+                call: ToolCall {
+                    id: "toolu_02DEF".to_string(),
+                    name: ToolName::Read,
+                    input: serde_json::json!({}),
+                    source_message_uuid: "msg-3".to_string(),
+                },
+                result: None,
+            },
+        ];
+
+        link_subagents_to_calls(&mut subagents, &exchanges);
+
+        assert_eq!(subagents[0].parent_tool_call_id, Some("toolu_01ABC".to_string()));
+        assert_eq!(subagents[1].parent_tool_call_id, None);
+    }
+
+    #[test]
+    fn test_link_subagents_empty_inputs() {
+        let mut empty_subagents: Vec<SubAgentSession> = vec![];
+        let empty_exchanges: Vec<ToolExchange> = vec![];
+        link_subagents_to_calls(&mut empty_subagents, &empty_exchanges);
+        // No panic
+    }
+
+    #[test]
+    fn test_link_subagents_no_match() {
+        let mut subagents = vec![SubAgentSession {
+            agent_id: "no_match".to_string(),
+            meta: SubAgentMeta { agent_type: None },
+            messages: vec![],
+            parent_tool_call_id: None,
+        }];
+
+        let exchanges = vec![ToolExchange {
+            call: ToolCall {
+                id: "toolu_99".to_string(),
+                name: ToolName::Agent,
+                input: serde_json::json!({}),
+                source_message_uuid: "msg-1".to_string(),
+            },
+            result: Some(ToolResultData {
+                tool_use_id: "toolu_99".to_string(),
+                content: Some("agentId: different_id\n<usage>tokens: 500</usage>".to_string()),
+                is_error: false,
+                source_message_uuid: "msg-2".to_string(),
+            }),
+        }];
+
+        link_subagents_to_calls(&mut subagents, &exchanges);
+        assert_eq!(subagents[0].parent_tool_call_id, None);
+    }
+}

crates/dirigent_anth/src/tree.rs

@@ -0,0 +1,171 @@
+//! Conversation tree module — builds a parent/child tree from `RawMessage`s.
+//!
+//! Claude Code sessions are not purely linear: the user can edit earlier
+//! messages, producing branches. Each message carries a `uuid` and a
+//! `parentUuid` that describe the relationship. This module reconstructs
+//! the tree so callers can walk threads, detect branches, and select the
+//! main thread.
+
+use std::collections::HashMap;
+
+use crate::types::RawMessage;
+
+// ---------------------------------------------------------------------------
+// Node & tree types
+// ---------------------------------------------------------------------------
+
+/// A single node in the conversation tree.
+#[derive(Debug)]
+pub struct ConversationNode {
+    /// The UUID of this message.
+    pub uuid: String,
+    /// The raw message stored at this node.
+    pub message: RawMessage,
+    /// UUIDs of direct children, in insertion order.
+    pub children: Vec<String>,
+}
+
+/// The full conversation tree for a session.
+///
+/// A session may have multiple roots when the first message has no
+/// `parentUuid`, or when a message refers to a parent that is not present
+/// in the slice provided to [`ConversationTree::build`].
+#[derive(Debug)]
+pub struct ConversationTree {
+    /// Root node UUIDs (messages with no parent or with an unknown parent).
+    pub roots: Vec<String>,
+    /// All nodes indexed by UUID.
+    pub nodes: HashMap<String, ConversationNode>,
+}
+
+// ---------------------------------------------------------------------------
+// UUID / parent-UUID helpers
+// ---------------------------------------------------------------------------
+
+/// Extract the `uuid` from any `RawMessage` variant.
+///
+/// Returns `None` for variants that carry no UUID (e.g. `QueueOperation`).
+pub fn message_uuid(msg: &RawMessage) -> Option<&str> {
+    match msg {
+        RawMessage::User(m) => m.uuid.as_deref(),
+        RawMessage::Assistant(m) => m.uuid.as_deref(),
+        RawMessage::Progress(m) => m.uuid.as_deref(),
+        RawMessage::System(m) => m.uuid.as_deref(),
+        RawMessage::QueueOperation(_)
+        | RawMessage::FileHistorySnapshot(_)
+        | RawMessage::LastPrompt(_) => None,
+    }
+}
+
+/// Extract the `parent_uuid` from any `RawMessage` variant.
+///
+/// Returns `None` for variants that carry no parent UUID.
+pub fn message_parent_uuid(msg: &RawMessage) -> Option<&str> {
+    match msg {
+        RawMessage::User(m) => m.parent_uuid.as_deref(),
+        RawMessage::Assistant(m) => m.parent_uuid.as_deref(),
+        RawMessage::Progress(m) => m.parent_uuid.as_deref(),
+        RawMessage::System(m) => m.parent_uuid.as_deref(),
+        RawMessage::QueueOperation(_)
+        | RawMessage::FileHistorySnapshot(_)
+        | RawMessage::LastPrompt(_) => None,
+    }
+}
+
+// ---------------------------------------------------------------------------
+// ConversationTree impl
+// ---------------------------------------------------------------------------
+
+impl ConversationTree {
+    /// Build a conversation tree from a sequence of messages.
+    ///
+    /// Messages without a UUID (e.g. `QueueOperation`) are silently skipped.
+    /// If a message's `parentUuid` is present but not found in the set,
+    /// that message is treated as a root.
+    pub fn build(messages: &[RawMessage]) -> Self {
+        let mut nodes: HashMap<String, ConversationNode> = HashMap::new();
+        let mut roots: Vec<String> = Vec::new();
+
+        // First pass: insert every addressable message as a node.
+        for msg in messages {
+            if let Some(uuid) = message_uuid(msg) {
+                nodes.insert(
+                    uuid.to_string(),
+                    ConversationNode {
+                        uuid: uuid.to_string(),
+                        message: msg.clone(),
+                        children: Vec::new(),
+                    },
+                );
+            }
+        }
+
+        // Second pass: collect (uuid, parent_uuid) pairs so we can wire up
+        // parent→child edges without a simultaneous mutable borrow.
+        let parent_links: Vec<(String, Option<String>)> = messages
+            .iter()
+            .filter_map(|msg| {
+                let uuid = message_uuid(msg)?.to_string();
+                let parent = message_parent_uuid(msg).map(|s| s.to_string());
+                Some((uuid, parent))
+            })
+            .collect();
+
+        for (uuid, parent_uuid) in parent_links {
+            match parent_uuid {
+                Some(parent_id) if nodes.contains_key(&parent_id) => {
+                    // Safe: parent_id != uuid (a message cannot be its own parent).
+                    nodes
+                        .get_mut(&parent_id)
+                        .expect("parent key confirmed above")
+                        .children
+                        .push(uuid);
+                }
+                _ => {
+                    // No parent, or parent not in the provided slice — treat as root.
+                    roots.push(uuid);
+                }
+            }
+        }
+
+        ConversationTree { roots, nodes }
+    }
+
+    /// Walk the *main thread*: start from the first root and always follow
+    /// the first child at each step.
+    ///
+    /// In a linear session this is the complete conversation. In a branching
+    /// session this is the path taken before any edits.
+    pub fn main_thread(&self) -> Vec<&ConversationNode> {
+        let mut result = Vec::new();
+        if let Some(root_id) = self.roots.first() {
+            let mut current = root_id.as_str();
+            loop {
+                match self.nodes.get(current) {
+                    Some(node) => {
+                        result.push(node);
+                        match node.children.first() {
+                            Some(first_child) => current = first_child.as_str(),
+                            None => break,
+                        }
+                    }
+                    None => break,
+                }
+            }
+        }
+        result
+    }
+
+    /// Returns `true` when every node has at most one child (no branches).
+    pub fn is_linear(&self) -> bool {
+        self.nodes.values().all(|n| n.children.len() <= 1)
+    }
+
+    /// Returns all nodes that have more than one child (branch points).
+    pub fn branch_points(&self) -> Vec<&ConversationNode> {
+        self.nodes
+            .values()
+            .filter(|n| n.children.len() > 1)
+            .collect()
+    }
+}

crates/dirigent_anth/src/types.rs

@@ -0,0 +1,847 @@
+//! Core types for parsing Claude Code JSONL session data.
+
+use camino::Utf8PathBuf;
+use serde::{Deserialize, Serialize};
+
+// ---------------------------------------------------------------------------
+// Content types
+// ---------------------------------------------------------------------------
+
+/// Content is either a plain string or an array of content blocks.
+///
+/// Uses a custom deserializer so that `Blocks` variant applies lenient
+/// deserialization — unknown content block types (e.g. `tool_reference`)
+/// are silently skipped instead of failing the entire message.
+#[derive(Debug, Clone, Serialize)]
+#[serde(untagged)]
+pub enum Content {
+    Text(String),
+    Blocks(Vec<ContentBlock>),
+}
+
+impl<'de> serde::Deserialize<'de> for Content {
+    fn deserialize<D>(deserializer: D) -> std::result::Result<Self, D::Error>
+    where
+        D: serde::Deserializer<'de>,
+    {
+        let value = serde_json::Value::deserialize(deserializer)?;
+        match value {
+            serde_json::Value::String(s) => Ok(Content::Text(s)),
+            serde_json::Value::Array(arr) => {
+                let blocks = arr
+                    .into_iter()
+                    .filter_map(|v| {
+                        serde_json::from_value::<ContentBlock>(v.clone())
+                            .ok()
+                            .or_else(|| {
+                                tracing::debug!(
+                                    "Skipping unknown content block: {:?}",
+                                    v.get("type")
+                                );
+                                None
+                            })
+                    })
+                    .collect();
+                Ok(Content::Blocks(blocks))
+            }
+            other => Err(serde::de::Error::custom(format!(
+                "expected string or array for Content, got {}",
+                match &other {
+                    serde_json::Value::Null => "null",
+                    serde_json::Value::Bool(_) => "bool",
+                    serde_json::Value::Number(_) => "number",
+                    serde_json::Value::Object(_) => "object",
+                    _ => "unknown",
+                }
+            ))),
+        }
+    }
+}
+
+/// Typed content block inside messages.
+#[derive(Debug, Clone, Deserialize, Serialize)]
+#[serde(tag = "type", rename_all = "snake_case")]
+pub enum ContentBlock {
+    Text {
+        text: String,
+    },
+    ToolUse {
+        id: String,
+        name: String,
+        input: serde_json::Value,
+        #[serde(default)]
+        caller: Option<serde_json::Value>,
+    },
+    ToolResult {
+        tool_use_id: String,
+        #[serde(default)]
+        content: Option<Content>,
+        #[serde(default)]
+        is_error: bool,
+    },
+    Thinking {
+        thinking: String,
+    },
+    Image {
+        source: serde_json::Value,
+    },
+}
+
+// ---------------------------------------------------------------------------
+// Lenient content block deserialization
+// ---------------------------------------------------------------------------
+
+/// Deserializes a `Vec<ContentBlock>` leniently — unknown block types are
+/// silently skipped instead of failing the entire message.
+fn deserialize_content_blocks<'de, D>(
+    deserializer: D,
+) -> std::result::Result<Vec<ContentBlock>, D::Error>
+where
+    D: serde::Deserializer<'de>,
+{
+    use serde::Deserialize as _;
+    let raw: Vec<serde_json::Value> = Vec::deserialize(deserializer)?;
+    Ok(raw
+        .into_iter()
+        .filter_map(|v| {
+            serde_json::from_value::<ContentBlock>(v.clone()).ok().or_else(|| {
+                tracing::debug!("Skipping unknown content block: {:?}", v.get("type"));
+                None
+            })
+        })
+        .collect())
+}
+
+// ---------------------------------------------------------------------------
+// Top-level JSONL line discriminator
+// ---------------------------------------------------------------------------
+
+/// Top-level JSONL line discriminator.
+#[derive(Debug, Clone, Deserialize)]
+#[serde(tag = "type", rename_all = "kebab-case")]
+pub enum RawMessage {
+    User(RawUserMessage),
+    Assistant(RawAssistantMessage),
+    Progress(RawProgressMessage),
+    System(RawSystemMessage),
+    QueueOperation(RawQueueOperation),
+    FileHistorySnapshot(RawFileHistorySnapshot),
+    LastPrompt(RawLastPrompt),
+}
+
+// ---------------------------------------------------------------------------
+// User message
+// ---------------------------------------------------------------------------
+
+#[derive(Debug, Clone, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct RawUserMessage {
+    #[serde(default)]
+    pub uuid: Option<String>,
+    #[serde(default)]
+    pub parent_uuid: Option<String>,
+    #[serde(default)]
+    pub timestamp: Option<String>,
+    #[serde(default)]
+    pub session_id: Option<String>,
+    #[serde(default)]
+    pub cwd: Option<String>,
+    #[serde(default)]
+    pub version: Option<String>,
+    #[serde(default)]
+    pub git_branch: Option<String>,
+    #[serde(default)]
+    pub is_sidechain: bool,
+    #[serde(default)]
+    pub is_meta: Option<bool>,
+    #[serde(default)]
+    pub user_type: Option<String>,
+    pub message: UserMessageInner,
+}
+
+#[derive(Debug, Clone, Deserialize, Serialize)]
+pub struct UserMessageInner {
+    pub role: String,
+    pub content: Content,
+}
+
+// ---------------------------------------------------------------------------
+// Assistant message
+// ---------------------------------------------------------------------------
+
+#[derive(Debug, Clone, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct RawAssistantMessage {
+    #[serde(default)]
+    pub uuid: Option<String>,
+    #[serde(default)]
+    pub parent_uuid: Option<String>,
+    #[serde(default)]
+    pub timestamp: Option<String>,
+    #[serde(default)]
+    pub session_id: Option<String>,
+    #[serde(default)]
+    pub cwd: Option<String>,
+    #[serde(default)]
+    pub version: Option<String>,
+    #[serde(default)]
+    pub git_branch: Option<String>,
+    #[serde(default)]
+    pub is_sidechain: bool,
+    #[serde(default)]
+    pub request_id: Option<String>,
+    pub message: AssistantInner,
+}
+
+// NOTE: AssistantInner is the Anthropic API response object nested inside
+// the Claude Code JSONL wrapper. The API uses snake_case (stop_reason, etc.)
+// unlike the outer JSONL wrapper which uses camelCase.
+#[derive(Debug, Clone, Deserialize)]
+pub struct AssistantInner {
+    #[serde(default)]
+    pub model: Option<String>,
+    #[serde(default)]
+    pub id: Option<String>,
+    #[serde(default, rename = "type")]
+    pub message_type: Option<String>,
+    #[serde(default)]
+    pub role: Option<String>,
+    #[serde(default, deserialize_with = "deserialize_content_blocks")]
+    pub content: Vec<ContentBlock>,
+    #[serde(default)]
+    pub stop_reason: Option<String>,
+    #[serde(default)]
+    pub stop_sequence: Option<String>,
+    #[serde(default)]
+    pub usage: Option<serde_json::Value>,
+}
+
+// ---------------------------------------------------------------------------
+// Progress message
+// ---------------------------------------------------------------------------
+
+#[derive(Debug, Clone, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct RawProgressMessage {
+    #[serde(default)]
+    pub uuid: Option<String>,
+    #[serde(default)]
+    pub parent_uuid: Option<String>,
+    #[serde(default)]
+    pub timestamp: Option<String>,
+    #[serde(default)]
+    pub session_id: Option<String>,
+    #[serde(default)]
+    pub cwd: Option<String>,
+    #[serde(default)]
+    pub version: Option<String>,
+    #[serde(default)]
+    pub git_branch: Option<String>,
+    #[serde(default)]
+    pub is_sidechain: bool,
+    #[serde(default)]
+    pub data: Option<serde_json::Value>,
+}
+
+// ---------------------------------------------------------------------------
+// System message
+// ---------------------------------------------------------------------------
+
+#[derive(Debug, Clone, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct RawSystemMessage {
+    #[serde(default)]
+    pub uuid: Option<String>,
+    #[serde(default)]
+    pub parent_uuid: Option<String>,
+    #[serde(default)]
+    pub timestamp: Option<String>,
+    #[serde(default)]
+    pub session_id: Option<String>,
+    #[serde(default)]
+    pub cwd: Option<String>,
+    #[serde(default)]
+    pub version: Option<String>,
+    #[serde(default)]
+    pub git_branch: Option<String>,
+    #[serde(default)]
+    pub is_sidechain: bool,
+    #[serde(default)]
+    pub data: Option<serde_json::Value>,
+}
+
+// ---------------------------------------------------------------------------
+// Queue operation
+// ---------------------------------------------------------------------------
+
+#[derive(Debug, Clone, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct RawQueueOperation {
+    pub operation: String,
+    #[serde(default)]
+    pub timestamp: Option<String>,
+    #[serde(default)]
+    pub session_id: Option<String>,
+}
+
+// ---------------------------------------------------------------------------
+// File history snapshot
+// ---------------------------------------------------------------------------
+
+#[derive(Debug, Clone, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct RawFileHistorySnapshot {
+    #[serde(default)]
+    pub message_id: Option<String>,
+    #[serde(default)]
+    pub is_snapshot_update: bool,
+    #[serde(default)]
+    pub snapshot: Option<serde_json::Value>,
+}
+
+// ---------------------------------------------------------------------------
+// Last prompt
+// ---------------------------------------------------------------------------
+
+#[derive(Debug, Clone, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct RawLastPrompt {
+    #[serde(default)]
+    pub last_prompt: Option<String>,
+    #[serde(default)]
+    pub session_id: Option<String>,
+}
+
+// ---------------------------------------------------------------------------
+// Tool types (for correlation module later)
+// ---------------------------------------------------------------------------
+
+/// Known tool names used by Claude Code.
+#[derive(Debug, Clone, PartialEq, Eq, Hash)]
+pub enum ToolName {
+    Bash,
+    Read,
+    Write,
+    Edit,
+    Grep,
+    Glob,
+    Agent,
+    Skill,
+    WebSearch,
+    WebFetch,
+    TodoWrite,
+    NotebookEdit,
+    Other(String),
+}
+
+impl From<String> for ToolName {
+    fn from(s: String) -> Self {
+        match s.as_str() {
+            "Bash" => ToolName::Bash,
+            "Read" => ToolName::Read,
+            "Write" => ToolName::Write,
+            "Edit" => ToolName::Edit,
+            "Grep" => ToolName::Grep,
+            "Glob" => ToolName::Glob,
+            "Agent" => ToolName::Agent,
+            "Skill" => ToolName::Skill,
+            "WebSearch" => ToolName::WebSearch,
+            "WebFetch" => ToolName::WebFetch,
+            "TodoWrite" => ToolName::TodoWrite,
+            "NotebookEdit" => ToolName::NotebookEdit,
+            other => ToolName::Other(other.to_string()),
+        }
+    }
+}
+
+/// A tool call extracted from an assistant message.
+#[derive(Debug, Clone)]
+pub struct ToolCall {
+    pub id: String,
+    pub name: ToolName,
+    pub input: serde_json::Value,
+    pub source_message_uuid: String,
+}
+
+/// A tool result extracted from a user message.
+#[derive(Debug, Clone)]
+pub struct ToolResultData {
+    pub tool_use_id: String,
+    pub content: Option<String>,
+    pub is_error: bool,
+    pub source_message_uuid: String,
+}
+
+/// A correlated tool call + result pair.
+#[derive(Debug, Clone)]
+pub struct ToolExchange {
+    pub call: ToolCall,
+    pub result: Option<ToolResultData>,
+}
+
+// ---------------------------------------------------------------------------
+// Discovery types (for discovery module later)
+// ---------------------------------------------------------------------------
+
+/// A discovered Claude Code project directory.
+#[derive(Debug, Clone)]
+pub struct ClaudeProject {
+    pub path: Utf8PathBuf,
+    pub original_path: String,
+    pub sessions: Vec<SessionRef>,
+}
+
+/// Reference to a session (not yet parsed).
+#[derive(Debug, Clone)]
+pub struct SessionRef {
+    pub id: String,
+    pub jsonl_path: Utf8PathBuf,
+    pub artifacts_dir: Option<Utf8PathBuf>,
+    pub index_entry: Option<SessionIndexEntry>,
+}
+
+/// From sessions-index.json (when available).
+#[derive(Debug, Clone, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct SessionIndexEntry {
+    #[serde(default)]
+    pub session_id: Option<String>,
+    #[serde(default)]
+    pub first_prompt: Option<String>,
+    #[serde(default)]
+    pub summary: Option<String>,
+    #[serde(default)]
+    pub message_count: Option<u32>,
+    #[serde(default)]
+    pub created: Option<serde_json::Value>,
+    #[serde(default)]
+    pub modified: Option<serde_json::Value>,
+    #[serde(default)]
+    pub git_branch: Option<String>,
+    #[serde(default)]
+    pub project_path: Option<String>,
+}
+
+// ---------------------------------------------------------------------------
+// Sub-agent types
+// ---------------------------------------------------------------------------
+
+/// Sub-agent metadata from .meta.json.
+#[derive(Debug, Clone, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct SubAgentMeta {
+    #[serde(default)]
+    pub agent_type: Option<String>,
+}
+
+/// A parsed sub-agent session.
+#[derive(Debug, Clone)]
+pub struct SubAgentSession {
+    pub agent_id: String,
+    pub meta: SubAgentMeta,
+    pub messages: Vec<RawMessage>,
+    pub parent_tool_call_id: Option<String>,
+}
+
+// ---------------------------------------------------------------------------
+// MessageMeta (convenience, future use)
+// ---------------------------------------------------------------------------
+
+/// Common metadata extracted from any message. Defined for future consumers.
+#[derive(Debug, Clone)]
+pub struct MessageMeta {
+    pub uuid: String,
+    pub parent_uuid: Option<String>,
+    pub timestamp: Option<String>,
+    pub session_id: String,
+    pub cwd: Option<String>,
+    pub version: Option<String>,
+    pub git_branch: Option<String>,
+    pub is_sidechain: bool,
+}
+
+// ---------------------------------------------------------------------------
+// ParsedSession
+// ---------------------------------------------------------------------------
+
+/// A fully parsed session with all correlations built.
+#[derive(Debug)]
+pub struct ParsedSession {
+    pub messages: Vec<RawMessage>,
+    pub tree: crate::tree::ConversationTree,
+    pub tool_exchanges: Vec<ToolExchange>,
+    pub subagents: Vec<SubAgentSession>,
+}
+
+// ---------------------------------------------------------------------------
+// Tests
+// ---------------------------------------------------------------------------
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn parse_content_text_string() {
+        let json = r#""Hello world""#;
+        let content: Content = serde_json::from_str(json).unwrap();
+        match content {
+            Content::Text(s) => assert_eq!(s, "Hello world"),
+            _ => panic!("Expected Content::Text"),
+        }
+    }
+
+    #[test]
+    fn parse_content_blocks() {
+        let json = r#"[{"type": "text", "text": "Hello"}]"#;
+        let content: Content = serde_json::from_str(json).unwrap();
+        match content {
+            Content::Blocks(blocks) => {
+                assert_eq!(blocks.len(), 1);
+                match &blocks[0] {
+                    ContentBlock::Text { text } => assert_eq!(text, "Hello"),
+                    _ => panic!("Expected ContentBlock::Text"),
+                }
+            }
+            _ => panic!("Expected Content::Blocks"),
+        }
+    }
+
+    #[test]
+    fn parse_tool_use_block() {
+        let json = r#"{"type": "tool_use", "id": "toolu_123", "name": "Bash", "input": {"command": "ls"}}"#;
+        let block: ContentBlock = serde_json::from_str(json).unwrap();
+        match block {
+            ContentBlock::ToolUse { id, name, .. } => {
+                assert_eq!(id, "toolu_123");
+                assert_eq!(name, "Bash");
+            }
+            _ => panic!("Expected ContentBlock::ToolUse"),
+        }
+    }
+
+    #[test]
+    fn parse_tool_result_block() {
+        let json = r#"{"type": "tool_result", "tool_use_id": "toolu_123", "content": "output text", "is_error": false}"#;
+        let block: ContentBlock = serde_json::from_str(json).unwrap();
+        match block {
+            ContentBlock::ToolResult {
+                tool_use_id,
+                is_error,
+                ..
+            } => {
+                assert_eq!(tool_use_id, "toolu_123");
+                assert!(!is_error);
+            }
+            _ => panic!("Expected ContentBlock::ToolResult"),
+        }
+    }
+
+    #[test]
+    fn parse_thinking_block() {
+        let json = r#"{"type": "thinking", "thinking": "Let me consider..."}"#;
+        let block: ContentBlock = serde_json::from_str(json).unwrap();
+        match block {
+            ContentBlock::Thinking { thinking } => {
+                assert_eq!(thinking, "Let me consider...");
+            }
+            _ => panic!("Expected ContentBlock::Thinking"),
+        }
+    }
+
+    #[test]
+    fn parse_queue_operation() {
+        let json = r#"{"type": "queue-operation", "operation": "enqueue", "timestamp": "2026-03-14T21:15:17.531Z", "sessionId": "00f72d8d-fc54-485c-a082-310ffcabdb73"}"#;
+        let msg: RawMessage = serde_json::from_str(json).unwrap();
+        match msg {
+            RawMessage::QueueOperation(op) => {
+                assert_eq!(op.operation, "enqueue");
+                assert_eq!(
+                    op.session_id.as_deref(),
+                    Some("00f72d8d-fc54-485c-a082-310ffcabdb73")
+                );
+            }
+            _ => panic!("Expected RawMessage::QueueOperation"),
+        }
+    }
+
+    #[test]
+    fn parse_user_message_with_string_content() {
+        let json = r#"{
+            "parentUuid": "b1ab1ac7-fdb6-4e25-bc17-4c060b470b4a",
+            "isSidechain": false,
+            "userType": "external",
+            "cwd": "G:\\dev\\projects\\dirigent",
+            "sessionId": "00f72d8d-fc54-485c-a082-310ffcabdb73",
+            "version": "2.1.71",
+            "gitBranch": "main",
+            "type": "user",
+            "message": {
+                "role": "user",
+                "content": "Hello world"
+            },
+            "isMeta": false,
+            "uuid": "1d843a4a-b99d-4c02-91a3-7cfe3dcac9f0",
+            "timestamp": "2026-03-14T21:08:58.586Z"
+        }"#;
+        let msg: RawMessage = serde_json::from_str(json).unwrap();
+        match msg {
+            RawMessage::User(u) => {
+                assert_eq!(u.uuid.as_deref(), Some("1d843a4a-b99d-4c02-91a3-7cfe3dcac9f0"));
+                assert_eq!(u.session_id.as_deref(), Some("00f72d8d-fc54-485c-a082-310ffcabdb73"));
+                assert_eq!(u.is_meta, Some(false));
+                match &u.message.content {
+                    Content::Text(s) => assert_eq!(s, "Hello world"),
+                    _ => panic!("Expected Content::Text"),
+                }
+            }
+            _ => panic!("Expected RawMessage::User"),
+        }
+    }
+
+    #[test]
+    fn parse_assistant_message_with_tool_use() {
+        let json = r#"{
+            "parentUuid": "77793647-f957-4aec-8b04-a9c07e01e37b",
+            "isSidechain": false,
+            "userType": "external",
+            "cwd": "G:\\dev\\projects\\dirigent",
+            "sessionId": "00f72d8d-fc54-485c-a082-310ffcabdb73",
+            "version": "2.1.71",
+            "gitBranch": "main",
+            "message": {
+                "model": "claude-opus-4-6",
+                "id": "msg_01NcwYjEydGEyZCSCgwmcnYd",
+                "type": "message",
+                "role": "assistant",
+                "content": [
+                    {
+                        "type": "tool_use",
+                        "id": "toolu_01DP5mkAQnAi2o54idq24cPn",
+                        "name": "Agent",
+                        "input": {
+                            "description": "Investigate config sources of truth",
+                            "subagent_type": "Explore",
+                            "prompt": "test prompt"
+                        },
+                        "caller": { "type": "direct" }
+                    }
+                ],
+                "stop_reason": null,
+                "stop_sequence": null,
+                "usage": {
+                    "input_tokens": 3,
+                    "cache_creation_input_tokens": 20147,
+                    "cache_read_input_tokens": 0,
+                    "output_tokens": 9,
+                    "service_tier": "standard"
+                }
+            },
+            "requestId": "req_011CZ3fYWGjcQCgh5d58d2k8",
+            "type": "assistant",
+            "uuid": "6cad0d13-d0ae-47fa-a6b1-b7b45a2b5e0b",
+            "timestamp": "2026-03-14T21:15:27.916Z"
+        }"#;
+        let msg: RawMessage = serde_json::from_str(json).unwrap();
+        match msg {
+            RawMessage::Assistant(a) => {
+                assert_eq!(a.uuid.as_deref(), Some("6cad0d13-d0ae-47fa-a6b1-b7b45a2b5e0b"));
+                assert_eq!(a.message.model.as_deref(), Some("claude-opus-4-6"));
+                assert_eq!(a.message.content.len(), 1);
+                match &a.message.content[0] {
+                    ContentBlock::ToolUse { name, id, .. } => {
+                        assert_eq!(name, "Agent");
+                        assert_eq!(id, "toolu_01DP5mkAQnAi2o54idq24cPn");
+                    }
+                    _ => panic!("Expected ContentBlock::ToolUse"),
+                }
+                assert!(a.message.stop_reason.is_none());
+                assert!(a.message.usage.is_some());
+            }
+            _ => panic!("Expected RawMessage::Assistant"),
+        }
+    }
+
+    #[test]
+    fn unknown_content_block_type_skipped_in_assistant() {
+        let json = r#"{
+            "parentUuid": null,
+            "isSidechain": false,
+            "sessionId": "test",
+            "message": {
+                "role": "assistant",
+                "content": [
+                    {"type": "text", "text": "known"},
+                    {"type": "future_type", "data": "something"}
+                ]
+            },
+            "type": "assistant",
+            "uuid": "test-uuid",
+            "timestamp": "2026-01-01T00:00:00Z"
+        }"#;
+        let msg: RawMessage = serde_json::from_str(json).unwrap();
+        match msg {
+            RawMessage::Assistant(a) => {
+                assert_eq!(a.message.content.len(), 1);
+                match &a.message.content[0] {
+                    ContentBlock::Text { text } => assert_eq!(text, "known"),
+                    _ => panic!("Expected ContentBlock::Text"),
+                }
+            }
+            _ => panic!("Expected RawMessage::Assistant"),
+        }
+    }
+
+    // -----------------------------------------------------------------------
+    // Regression tests for parse failure audit (2026-04-04)
+    // -----------------------------------------------------------------------
+
+    #[test]
+    fn tool_reference_in_tool_result_content_does_not_fail() {
+        // Suggestion 1 & 3: tool_reference blocks inside tool_result.content
+        // should be silently skipped, not fail the entire message.
+        let json = r#"{
+            "type": "user",
+            "uuid": "test-uuid",
+            "parentUuid": null,
+            "isSidechain": false,
+            "sessionId": "test-session",
+            "message": {
+                "role": "user",
+                "content": [
+                    {
+                        "type": "tool_result",
+                        "tool_use_id": "toolu_abc123",
+                        "content": [
+                            {"type": "text", "text": "File contents here"},
+                            {"type": "tool_reference", "tool_name": "TodoWrite"}
+                        ],
+                        "is_error": false
+                    }
+                ]
+            }
+        }"#;
+        let msg: RawMessage = serde_json::from_str(json).unwrap();
+        match msg {
+            RawMessage::User(u) => {
+                match &u.message.content {
+                    Content::Blocks(blocks) => {
+                        assert_eq!(blocks.len(), 1);
+                        match &blocks[0] {
+                            ContentBlock::ToolResult { tool_use_id, content, .. } => {
+                                assert_eq!(tool_use_id, "toolu_abc123");
+                                // The inner content should have 1 block (text), tool_reference skipped
+                                match content.as_ref().unwrap() {
+                                    Content::Blocks(inner) => {
+                                        assert_eq!(inner.len(), 1);
+                                        match &inner[0] {
+                                            ContentBlock::Text { text } => {
+                                                assert_eq!(text, "File contents here");
+                                            }
+                                            _ => panic!("Expected inner ContentBlock::Text"),
+                                        }
+                                    }
+                                    _ => panic!("Expected inner Content::Blocks"),
+                                }
+                            }
+                            _ => panic!("Expected ContentBlock::ToolResult"),
+                        }
+                    }
+                    _ => panic!("Expected Content::Blocks"),
+                }
+            }
+            _ => panic!("Expected RawMessage::User"),
+        }
+    }
+
+    #[test]
+    fn file_history_snapshot_parses() {
+        // Suggestion 2: file-history-snapshot lines should parse, not fail.
+        let json = r#"{
+            "type": "file-history-snapshot",
+            "messageId": "abc-123",
+            "isSnapshotUpdate": false,
+            "snapshot": {
+                "messageId": "abc-123",
+                "trackedFileBackups": {
+                    "src/main.rs": {"backupFileName": "main.rs.bak", "backupTime": "2026-01-01T00:00:00Z", "version": "1"}
+                },
+                "timestamp": "2026-01-01T00:00:00Z"
+            }
+        }"#;
+        let msg: RawMessage = serde_json::from_str(json).unwrap();
+        match msg {
+            RawMessage::FileHistorySnapshot(s) => {
+                assert_eq!(s.message_id.as_deref(), Some("abc-123"));
+                assert!(!s.is_snapshot_update);
+                assert!(s.snapshot.is_some());
+            }
+            _ => panic!("Expected RawMessage::FileHistorySnapshot"),
+        }
+    }
+
+    #[test]
+    fn last_prompt_parses() {
+        // Suggestion 2: last-prompt lines should parse, not fail.
+        let json = r#"{
+            "type": "last-prompt",
+            "lastPrompt": "Fix the bug in auth middleware",
+            "sessionId": "session-uuid-123"
+        }"#;
+        let msg: RawMessage = serde_json::from_str(json).unwrap();
+        match msg {
+            RawMessage::LastPrompt(lp) => {
+                assert_eq!(lp.last_prompt.as_deref(), Some("Fix the bug in auth middleware"));
+                assert_eq!(lp.session_id.as_deref(), Some("session-uuid-123"));
+            }
+            _ => panic!("Expected RawMessage::LastPrompt"),
+        }
+    }
+
+    #[test]
+    fn unknown_content_block_in_user_message_skipped() {
+        // Suggestion 3: Unknown block types in user message content
+        // should be silently skipped (lenient everywhere).
+        let json = r#"{
+            "type": "user",
+            "uuid": "test-uuid",
+            "isSidechain": false,
+            "sessionId": "test",
+            "message": {
+                "role": "user",
+                "content": [
+                    {"type": "text", "text": "known"},
+                    {"type": "future_unknown_type", "data": "something"}
+                ]
+            }
+        }"#;
+        let msg: RawMessage = serde_json::from_str(json).unwrap();
+        match msg {
+            RawMessage::User(u) => {
+                match &u.message.content {
+                    Content::Blocks(blocks) => {
+                        assert_eq!(blocks.len(), 1);
+                        match &blocks[0] {
+                            ContentBlock::Text { text } => assert_eq!(text, "known"),
+                            _ => panic!("Expected ContentBlock::Text"),
+                        }
+                    }
+                    _ => panic!("Expected Content::Blocks"),
+                }
+            }
+            _ => panic!("Expected RawMessage::User"),
+        }
+    }
+
+    #[test]
+    fn tool_name_from_string() {
+        assert_eq!(ToolName::from("Bash".to_string()), ToolName::Bash);
+        assert_eq!(ToolName::from("Read".to_string()), ToolName::Read);
+        assert_eq!(ToolName::from("Agent".to_string()), ToolName::Agent);
+        assert_eq!(ToolName::from("WebSearch".to_string()), ToolName::WebSearch);
+        assert_eq!(
+            ToolName::from("CustomTool".to_string()),
+            ToolName::Other("CustomTool".to_string())
+        );
+    }
+}

crates/dirigent_anth/src/util.rs

@@ -0,0 +1,70 @@
+use chrono::{DateTime, Utc};
+
+/// Parse a timestamp from various formats found in Claude Code data.
+///
+/// Supports:
+/// - ISO 8601 string: "2026-03-22T17:00:13.192Z"
+/// - Unix milliseconds (number > 1e12): 1769461914249
+/// - Unix seconds (number <= 1e12): 1769461914
+pub fn parse_timestamp(value: &serde_json::Value) -> Option<DateTime<Utc>> {
+    match value {
+        serde_json::Value::String(s) => {
+            DateTime::parse_from_rfc3339(s)
+                .ok()
+                .map(|dt| dt.with_timezone(&Utc))
+        }
+        serde_json::Value::Number(n) => {
+            if let Some(ms) = n.as_i64() {
+                if ms > 1_000_000_000_000 {
+                    DateTime::from_timestamp_millis(ms)
+                } else {
+                    DateTime::from_timestamp(ms, 0)
+                }
+            } else {
+                None
+            }
+        }
+        _ => None,
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use chrono::Datelike;
+
+    #[test]
+    fn parse_timestamp_iso8601() {
+        let v = serde_json::json!("2026-03-22T17:00:13.192Z");
+        let dt = parse_timestamp(&v).unwrap();
+        assert_eq!(dt.year(), 2026);
+        assert_eq!(dt.month(), 3);
+        assert_eq!(dt.day(), 22);
+    }
+
+    #[test]
+    fn parse_timestamp_unix_millis() {
+        let v = serde_json::json!(1769461914249_i64);
+        let dt = parse_timestamp(&v).unwrap();
+        assert!(dt.year() >= 2025);
+    }
+
+    #[test]
+    fn parse_timestamp_unix_seconds() {
+        let v = serde_json::json!(1769461914_i64);
+        let dt = parse_timestamp(&v).unwrap();
+        assert!(dt.year() >= 2025);
+    }
+
+    #[test]
+    fn parse_timestamp_null_returns_none() {
+        let v = serde_json::json!(null);
+        assert!(parse_timestamp(&v).is_none());
+    }
+
+    #[test]
+    fn parse_timestamp_invalid_string_returns_none() {
+        let v = serde_json::json!("not a date");
+        assert!(parse_timestamp(&v).is_none());
+    }
+}

crates/dirigent_anth/tests/fixtures/branching_tree.jsonl

@@ -0,0 +1,6 @@
+{"type":"user","uuid":"r-001","parentUuid":null,"timestamp":"2026-03-23T10:00:00.000Z","sessionId":"test-session-tree","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"isMeta":false,"userType":"external","message":{"role":"user","content":"Help me"}}
+{"type":"assistant","uuid":"a-001","parentUuid":"r-001","timestamp":"2026-03-23T10:00:01.000Z","sessionId":"test-session-tree","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"requestId":"req-001","message":{"model":"claude-opus-4-6","id":"msg-001","type":"message","role":"assistant","content":[{"type":"text","text":"Sure"}],"stop_reason":"end_turn","usage":{"input_tokens":10,"output_tokens":5}}}
+{"type":"user","uuid":"u-002","parentUuid":"a-001","timestamp":"2026-03-23T10:00:02.000Z","sessionId":"test-session-tree","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"isMeta":false,"userType":"external","message":{"role":"user","content":"Do option A"}}
+{"type":"assistant","uuid":"a-003","parentUuid":"u-002","timestamp":"2026-03-23T10:00:03.000Z","sessionId":"test-session-tree","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"requestId":"req-002","message":{"model":"claude-opus-4-6","id":"msg-003","type":"message","role":"assistant","content":[{"type":"text","text":"Doing A"}],"stop_reason":"end_turn","usage":{"input_tokens":15,"output_tokens":5}}}
+{"type":"user","uuid":"u-002b","parentUuid":"a-001","timestamp":"2026-03-23T10:00:04.000Z","sessionId":"test-session-tree","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"isMeta":false,"userType":"external","message":{"role":"user","content":"Actually, do option B"}}
+{"type":"assistant","uuid":"a-003b","parentUuid":"u-002b","timestamp":"2026-03-23T10:00:05.000Z","sessionId":"test-session-tree","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"requestId":"req-003","message":{"model":"claude-opus-4-6","id":"msg-003b","type":"message","role":"assistant","content":[{"type":"text","text":"Doing B"}],"stop_reason":"end_turn","usage":{"input_tokens":15,"output_tokens":5}}}

crates/dirigent_anth/tests/fixtures/minimal_session.jsonl

@@ -0,0 +1,6 @@
+{"type":"queue-operation","operation":"enqueue","timestamp":"2026-03-14T21:00:00.000Z","sessionId":"test-session-001"}
+{"type":"queue-operation","operation":"dequeue","timestamp":"2026-03-14T21:00:00.001Z","sessionId":"test-session-001"}
+{"type":"user","uuid":"u-001","parentUuid":null,"timestamp":"2026-03-14T21:00:01.000Z","sessionId":"test-session-001","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"isMeta":false,"userType":"external","message":{"role":"user","content":"Hello, help me with this project"}}
+{"type":"assistant","uuid":"a-001","parentUuid":"u-001","timestamp":"2026-03-14T21:00:02.000Z","sessionId":"test-session-001","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"requestId":"req-001","message":{"model":"claude-opus-4-6","id":"msg-001","type":"message","role":"assistant","content":[{"type":"text","text":"I'll help you."},{"type":"tool_use","id":"toolu_01","name":"Bash","input":{"command":"ls","description":"List files"}}],"stop_reason":"tool_use","usage":{"input_tokens":100,"output_tokens":50}}}
+{"type":"user","uuid":"u-002","parentUuid":"a-001","timestamp":"2026-03-14T21:00:03.000Z","sessionId":"test-session-001","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"isMeta":true,"userType":"external","message":{"role":"user","content":[{"type":"tool_result","tool_use_id":"toolu_01","content":"file1.rs\nfile2.rs","is_error":false}]}}
+{"type":"assistant","uuid":"a-002","parentUuid":"u-002","timestamp":"2026-03-14T21:00:04.000Z","sessionId":"test-session-001","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"requestId":"req-002","message":{"model":"claude-opus-4-6","id":"msg-002","type":"message","role":"assistant","content":[{"type":"text","text":"I can see two Rust files in the directory."}],"stop_reason":"end_turn","usage":{"input_tokens":200,"output_tokens":30}}}

crates/dirigent_anth/tests/fixtures/noise_patterns.jsonl

@@ -0,0 +1,9 @@
+{"type":"queue-operation","operation":"enqueue","timestamp":"2026-03-14T21:00:00.000Z","sessionId":"test-session-noise"}
+{"type":"user","uuid":"u-n-001","parentUuid":null,"timestamp":"2026-03-14T21:00:01.000Z","sessionId":"test-session-noise","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"isMeta":true,"message":{"role":"user","content":"system injected stuff"}}
+{"type":"user","uuid":"u-n-002","parentUuid":"u-n-001","timestamp":"2026-03-14T21:00:02.000Z","sessionId":"test-session-noise","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"isMeta":false,"message":{"role":"user","content":"Warmup"}}
+{"type":"user","uuid":"u-n-003","parentUuid":"u-n-002","timestamp":"2026-03-14T21:00:03.000Z","sessionId":"test-session-noise","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"isMeta":false,"message":{"role":"user","content":"[Request interrupted by user"}}
+{"type":"user","uuid":"u-n-004","parentUuid":"u-n-003","timestamp":"2026-03-14T21:00:04.000Z","sessionId":"test-session-noise","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"isMeta":false,"message":{"role":"user","content":"This session is being continued from a previous conversation"}}
+{"type":"user","uuid":"u-n-005","parentUuid":"u-n-004","timestamp":"2026-03-14T21:00:05.000Z","sessionId":"test-session-noise","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"isMeta":false,"message":{"role":"user","content":"API Error: rate limit exceeded"}}
+{"type":"user","uuid":"u-n-006","parentUuid":"u-n-005","timestamp":"2026-03-14T21:00:06.000Z","sessionId":"test-session-noise","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"isMeta":false,"message":{"role":"user","content":"Caveat: The messages below were generated by the user"}}
+{"type":"user","uuid":"u-n-007","parentUuid":"u-n-006","timestamp":"2026-03-14T21:00:07.000Z","sessionId":"test-session-noise","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"isMeta":false,"message":{"role":"user","content":"Please help me fix this bug"}}
+{"type":"assistant","uuid":"a-n-001","parentUuid":"u-n-007","timestamp":"2026-03-14T21:00:08.000Z","sessionId":"test-session-noise","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"message":{"id":"msg-n-001","role":"assistant","content":[{"type":"text","text":"Sure, let me help."}],"stop_reason":"end_turn"}}

crates/dirigent_anth/tests/fixtures/streaming_dedup.jsonl

@@ -0,0 +1,6 @@
+{"type":"user","uuid":"u-100","parentUuid":null,"timestamp":"2026-03-23T10:00:00.000Z","sessionId":"test-session-dedup","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"isMeta":false,"userType":"external","message":{"role":"user","content":"What files are here?"}}
+{"type":"assistant","uuid":"a-100","parentUuid":"u-100","timestamp":"2026-03-23T10:00:01.000Z","sessionId":"test-session-dedup","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"requestId":"req-100","message":{"model":"claude-opus-4-6","id":"msg-100","type":"message","role":"assistant","content":[{"type":"text","text":"Let me"}],"stop_reason":null,"usage":{"input_tokens":50,"output_tokens":3}}}
+{"type":"assistant","uuid":"a-100","parentUuid":"u-100","timestamp":"2026-03-23T10:00:01.100Z","sessionId":"test-session-dedup","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"requestId":"req-100","message":{"model":"claude-opus-4-6","id":"msg-100","type":"message","role":"assistant","content":[{"type":"text","text":"Let me look"},{"type":"tool_use","id":"toolu_100","name":"Bash","input":{"command":""}}],"stop_reason":null,"usage":{"input_tokens":50,"output_tokens":12}}}
+{"type":"assistant","uuid":"a-100","parentUuid":"u-100","timestamp":"2026-03-23T10:00:01.200Z","sessionId":"test-session-dedup","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"requestId":"req-100","message":{"model":"claude-opus-4-6","id":"msg-100","type":"message","role":"assistant","content":[{"type":"text","text":"Let me look at this."},{"type":"tool_use","id":"toolu_100","name":"Bash","input":{"command":"ls"}}],"stop_reason":"tool_use","usage":{"input_tokens":50,"output_tokens":20}}}
+{"type":"user","uuid":"u-101","parentUuid":"a-100","timestamp":"2026-03-23T10:00:02.000Z","sessionId":"test-session-dedup","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"isMeta":true,"userType":"external","message":{"role":"user","content":[{"type":"tool_result","tool_use_id":"toolu_100","content":"main.rs\nlib.rs","is_error":false}]}}
+{"type":"assistant","uuid":"a-101","parentUuid":"u-101","timestamp":"2026-03-23T10:00:03.000Z","sessionId":"test-session-dedup","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"requestId":"req-101","message":{"model":"claude-opus-4-6","id":"msg-101","type":"message","role":"assistant","content":[{"type":"text","text":"Done."}],"stop_reason":"end_turn","usage":{"input_tokens":100,"output_tokens":5}}}

crates/dirigent_anth/tests/fixtures/subagent/parent.jsonl

@@ -0,0 +1,4 @@
+{"type":"user","uuid":"u-300","parentUuid":null,"timestamp":"2026-03-23T12:00:00.000Z","sessionId":"test-session-sub","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"isMeta":false,"message":{"role":"user","content":"Search the codebase"}}
+{"type":"assistant","uuid":"a-300","parentUuid":"u-300","timestamp":"2026-03-23T12:00:01.000Z","sessionId":"test-session-sub","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"requestId":"req-300","message":{"model":"claude-opus-4-6","id":"msg-300","type":"message","role":"assistant","content":[{"type":"tool_use","id":"toolu_300","name":"Agent","input":{"description":"Search codebase","subagent_type":"Explore","prompt":"Find all config files"}}],"stop_reason":"tool_use","usage":{"input_tokens":100,"output_tokens":20}}}
+{"type":"user","uuid":"u-301","parentUuid":"a-300","timestamp":"2026-03-23T12:00:30.000Z","sessionId":"test-session-sub","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"isMeta":true,"message":{"role":"user","content":[{"type":"tool_result","tool_use_id":"toolu_300","content":"Found 3 config files","is_error":false}]}}
+{"type":"assistant","uuid":"a-301","parentUuid":"u-301","timestamp":"2026-03-23T12:00:31.000Z","sessionId":"test-session-sub","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"requestId":"req-301","message":{"model":"claude-opus-4-6","id":"msg-301","type":"message","role":"assistant","content":[{"type":"text","text":"I found the config files."}],"stop_reason":"end_turn","usage":{"input_tokens":200,"output_tokens":10}}}

crates/dirigent_anth/tests/fixtures/subagent/parent/subagents/agent-abc123.jsonl

@@ -0,0 +1,2 @@
+{"type":"user","uuid":"sa-u1","parentUuid":null,"timestamp":"2026-03-23T12:00:02.000Z","sessionId":"agent-abc123","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":true,"isMeta":false,"message":{"role":"user","content":"Find all config files"}}
+{"type":"assistant","uuid":"sa-a1","parentUuid":"sa-u1","timestamp":"2026-03-23T12:00:03.000Z","sessionId":"agent-abc123","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":true,"requestId":"req-sa1","message":{"model":"claude-opus-4-6","id":"msg-sa1","type":"message","role":"assistant","content":[{"type":"text","text":"Found config.toml, settings.json, .env"}],"stop_reason":"end_turn","usage":{"input_tokens":50,"output_tokens":15}}}

crates/dirigent_anth/tests/fixtures/subagent/parent/subagents/agent-abc123.meta.json

@@ -0,0 +1 @@
+{"agentType": "Explore"}

crates/dirigent_anth/tests/fixtures/tool_correlation.jsonl

@@ -0,0 +1,6 @@
+{"type":"user","uuid":"u-200","parentUuid":null,"timestamp":"2026-03-23T10:00:00.000Z","sessionId":"test-session-corr","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"isMeta":false,"userType":"external","message":{"role":"user","content":"Fix the bug"}}
+{"type":"assistant","uuid":"a-200","parentUuid":"u-200","timestamp":"2026-03-23T10:00:01.000Z","sessionId":"test-session-corr","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"requestId":"req-200","message":{"model":"claude-opus-4-6","id":"msg-200","type":"message","role":"assistant","content":[{"type":"tool_use","id":"toolu_200","name":"Bash","input":{"command":"cargo test"}},{"type":"tool_use","id":"toolu_201","name":"Read","input":{"file_path":"src/main.rs"}}],"stop_reason":"tool_use","usage":{"input_tokens":100,"output_tokens":50}}}
+{"type":"user","uuid":"u-201","parentUuid":"a-200","timestamp":"2026-03-23T10:00:02.000Z","sessionId":"test-session-corr","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"isMeta":true,"userType":"external","message":{"role":"user","content":[{"type":"tool_result","tool_use_id":"toolu_200","content":"test result output","is_error":false},{"type":"tool_result","tool_use_id":"toolu_201","content":"fn main() {}","is_error":false}]}}
+{"type":"assistant","uuid":"a-201","parentUuid":"u-201","timestamp":"2026-03-23T10:00:03.000Z","sessionId":"test-session-corr","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"requestId":"req-201","message":{"model":"claude-opus-4-6","id":"msg-201","type":"message","role":"assistant","content":[{"type":"tool_use","id":"toolu_202","name":"Write","input":{"file_path":"src/fix.rs","content":"fixed"}}],"stop_reason":"tool_use","usage":{"input_tokens":150,"output_tokens":30}}}
+{"type":"user","uuid":"u-202","parentUuid":"a-201","timestamp":"2026-03-23T10:00:04.000Z","sessionId":"test-session-corr","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"isMeta":true,"userType":"external","message":{"role":"user","content":[{"type":"tool_result","tool_use_id":"toolu_202","content":"File written successfully","is_error":false}]}}
+{"type":"assistant","uuid":"a-202","parentUuid":"u-202","timestamp":"2026-03-23T10:00:05.000Z","sessionId":"test-session-corr","cwd":"G:\\dev\\test","version":"2.1.71","gitBranch":"main","isSidechain":false,"requestId":"req-202","message":{"model":"claude-opus-4-6","id":"msg-202","type":"message","role":"assistant","content":[{"type":"text","text":"Bug is fixed."}],"stop_reason":"end_turn","usage":{"input_tokens":200,"output_tokens":20}}}

crates/dirigent_anth/tests/integration_tests.rs

@@ -0,0 +1,294 @@
+use camino::{Utf8Path, Utf8PathBuf};
+use chrono::Datelike;
+use dirigent_anth::{
+    correlation::correlate_tools,
+    dedup::dedup_messages,
+    noise::{classify_noise, NoiseKind},
+    parse_session,
+    tree::ConversationTree,
+    types::{ContentBlock, RawMessage},
+    util::parse_timestamp,
+};
+
+#[test]
+fn parse_minimal_session() {
+    let path = Utf8Path::new("tests/fixtures/minimal_session.jsonl");
+    let messages = parse_session(path).unwrap();
+
+    assert_eq!(messages.len(), 6, "Expected 6 messages, got {}", messages.len());
+
+    let type_names: Vec<&str> = messages
+        .iter()
+        .map(|m| match m {
+            RawMessage::User(_) => "user",
+            RawMessage::Assistant(_) => "assistant",
+            RawMessage::Progress(_) => "progress",
+            RawMessage::System(_) => "system",
+            RawMessage::QueueOperation(_) => "queue-operation",
+            RawMessage::FileHistorySnapshot(_) => "file-history-snapshot",
+            RawMessage::LastPrompt(_) => "last-prompt",
+        })
+        .collect();
+
+    assert_eq!(
+        type_names.iter().filter(|&&t| t == "queue-operation").count(),
+        2
+    );
+    assert_eq!(type_names.iter().filter(|&&t| t == "user").count(), 2);
+    assert_eq!(
+        type_names.iter().filter(|&&t| t == "assistant").count(),
+        2
+    );
+}
+
+#[test]
+fn parse_line_returns_none_for_invalid_json() {
+    assert!(dirigent_anth::parse_line("not valid json", 1).is_none());
+    assert!(dirigent_anth::parse_line("{}", 1).is_none());
+}
+
+#[test]
+fn dedup_streaming_session() {
+    let path = Utf8Path::new("tests/fixtures/streaming_dedup.jsonl");
+    let messages = parse_session(path).unwrap();
+
+    // Raw should have 6 lines (including 3 versions of same assistant message)
+    assert_eq!(messages.len(), 6, "Raw messages: expected 6, got {}", messages.len());
+
+    let deduped = dedup_messages(messages);
+
+    // After dedup: U1, A1(final), U2, A2 = 4
+    assert_eq!(deduped.len(), 4, "Deduped messages: expected 4, got {}", deduped.len());
+
+    // The kept assistant message must be the final version
+    let first_assistant = deduped.iter().find(|m| matches!(m, RawMessage::Assistant(_))).unwrap();
+    if let RawMessage::Assistant(a) = first_assistant {
+        assert!(a.message.stop_reason.is_some(), "Deduped assistant should have stop_reason set");
+        assert_eq!(a.message.stop_reason.as_deref(), Some("tool_use"));
+        assert_eq!(a.message.content.len(), 2, "Final version should have 2 content blocks");
+    } else {
+        unreachable!();
+    }
+}
+
+#[test]
+fn dedup_preserves_non_streamed_messages() {
+    let path = Utf8Path::new("tests/fixtures/minimal_session.jsonl");
+    let messages = parse_session(path).unwrap();
+    let count_before = messages.len();
+    let deduped = dedup_messages(messages);
+    // No streaming in minimal_session, so count should be same
+    assert_eq!(deduped.len(), count_before);
+}
+
+#[test]
+fn correlate_parallel_tools() {
+    let path = Utf8Path::new("tests/fixtures/tool_correlation.jsonl");
+    let messages = dirigent_anth::parse_session_deduped(path).unwrap();
+    let exchanges = correlate_tools(&messages);
+
+    // 3 tool calls: 2 parallel (Bash+Read) + 1 sequential (Write)
+    assert_eq!(exchanges.len(), 3);
+
+    // All should have results
+    assert!(exchanges.iter().all(|e| e.result.is_some()));
+
+    // Verify correct pairing by ID
+    for ex in &exchanges {
+        assert_eq!(ex.call.id, ex.result.as_ref().unwrap().tool_use_id);
+    }
+}
+
+#[test]
+fn correlate_no_tools_returns_empty() {
+    // Test with just a plain user message — no tool calls or results
+    let messages = vec![
+        serde_json::from_str::<RawMessage>(
+            r#"{"type":"user","uuid":"x","timestamp":"2026-01-01T00:00:00Z","sessionId":"s","message":{"role":"user","content":"hello"}}"#,
+        )
+        .unwrap(),
+    ];
+    let exchanges = correlate_tools(&messages);
+    assert!(exchanges.is_empty());
+}
+
+#[test]
+fn build_branching_tree() {
+    let path = Utf8Path::new("tests/fixtures/branching_tree.jsonl");
+    let messages = dirigent_anth::parse_session(path).unwrap();
+    let tree = ConversationTree::build(&messages);
+
+    assert_eq!(tree.roots.len(), 1);
+    assert!(!tree.is_linear());
+    assert_eq!(tree.branch_points().len(), 1); // A1 has 2 children
+
+    let main = tree.main_thread();
+    assert_eq!(main.len(), 4); // R → A1 → U2 → A3 (first branch)
+}
+
+#[test]
+fn linear_conversation_is_linear() {
+    let path = Utf8Path::new("tests/fixtures/minimal_session.jsonl");
+    let messages = dirigent_anth::parse_session(path).unwrap();
+    let tree = ConversationTree::build(&messages);
+    assert!(tree.is_linear());
+}
+
+#[test]
+fn classify_noise_from_fixture() {
+    let path = Utf8Path::new("tests/fixtures/noise_patterns.jsonl");
+    let messages = dirigent_anth::parse_session(path).unwrap();
+
+    assert_eq!(messages.len(), 9, "Expected 9 messages in noise fixture");
+
+    let classifications: Vec<Option<NoiseKind>> = messages.iter()
+        .map(classify_noise)
+        .collect();
+
+    assert_eq!(classifications[0], Some(NoiseKind::QueueOp));
+    assert_eq!(classifications[1], Some(NoiseKind::Meta));
+    assert_eq!(classifications[2], Some(NoiseKind::Warmup));
+    assert_eq!(classifications[3], Some(NoiseKind::Interrupted));
+    assert_eq!(classifications[4], Some(NoiseKind::Continuation));
+    assert_eq!(classifications[5], Some(NoiseKind::ApiError));
+    assert_eq!(classifications[6], Some(NoiseKind::SystemCaveat));
+    assert_eq!(classifications[7], None); // normal user
+    assert_eq!(classifications[8], None); // normal assistant
+}
+
+#[test]
+fn load_subagent_from_fixture() {
+    let artifacts_dir = Utf8Path::new("tests/fixtures/subagent/parent");
+    let subagents = dirigent_anth::load_subagents(artifacts_dir).unwrap();
+
+    assert_eq!(subagents.len(), 1);
+    assert_eq!(subagents[0].agent_id, "abc123");
+    assert_eq!(subagents[0].meta.agent_type.as_deref(), Some("Explore"));
+    assert_eq!(subagents[0].messages.len(), 2);
+}
+
+#[test]
+fn load_subagents_empty_dir() {
+    // Non-existent artifacts dir should return empty vec
+    let artifacts_dir = Utf8Path::new("tests/fixtures/nonexistent");
+    let subagents = dirigent_anth::load_subagents(artifacts_dir).unwrap();
+    assert!(subagents.is_empty());
+}
+
+#[test]
+fn load_full_session_with_subagents() {
+    use dirigent_anth::types::SessionRef;
+
+    let session_ref = SessionRef {
+        id: "parent".to_string(),
+        jsonl_path: Utf8PathBuf::from("tests/fixtures/subagent/parent.jsonl"),
+        artifacts_dir: Some(Utf8PathBuf::from("tests/fixtures/subagent/parent")),
+        index_entry: None,
+    };
+
+    let session = dirigent_anth::load_session(&session_ref).unwrap();
+    assert!(!session.messages.is_empty());
+    assert!(!session.subagents.is_empty());
+    assert!(!session.tree.roots.is_empty());
+    assert!(!session.tool_exchanges.is_empty());
+}
+
+#[test]
+fn load_session_without_artifacts() {
+    use dirigent_anth::types::SessionRef;
+
+    let session_ref = SessionRef {
+        id: "minimal".to_string(),
+        jsonl_path: Utf8PathBuf::from("tests/fixtures/minimal_session.jsonl"),
+        artifacts_dir: None,
+        index_entry: None,
+    };
+
+    let session = dirigent_anth::load_session(&session_ref).unwrap();
+    assert_eq!(session.messages.len(), 6); // 2 queue-ops + 2 users + 2 assistants
+    assert!(session.subagents.is_empty());
+    assert!(session.tree.is_linear());
+}
+
+#[test]
+fn content_as_string_or_blocks() {
+    // String content
+    let s: dirigent_anth::types::Content = serde_json::from_str(r#""hello""#).unwrap();
+    assert!(matches!(s, dirigent_anth::types::Content::Text(_)));
+
+    // Block content
+    let b: dirigent_anth::types::Content =
+        serde_json::from_str(r#"[{"type":"text","text":"hi"}]"#).unwrap();
+    assert!(matches!(b, dirigent_anth::types::Content::Blocks(_)));
+
+    // Empty blocks
+    let empty: dirigent_anth::types::Content = serde_json::from_str(r#"[]"#).unwrap();
+    assert!(matches!(empty, dirigent_anth::types::Content::Blocks(ref v) if v.is_empty()));
+}
+
+#[test]
+fn missing_optional_fields_dont_crash() {
+    // Minimal assistant message with many fields missing
+    let json = r#"{
+        "type": "assistant",
+        "message": {
+            "content": [{"type": "text", "text": "hi"}]
+        }
+    }"#;
+    let msg: RawMessage = serde_json::from_str(json).unwrap();
+    assert!(matches!(msg, RawMessage::Assistant(_)));
+}
+
+#[test]
+fn tool_result_content_string_and_blocks() {
+    // tool_result with string content
+    let json = r#"{"type":"tool_result","tool_use_id":"t1","content":"output text","is_error":false}"#;
+    let block: ContentBlock = serde_json::from_str(json).unwrap();
+    if let ContentBlock::ToolResult { content, is_error, .. } = block {
+        assert!(!is_error);
+        assert!(content.is_some());
+    } else {
+        panic!("Expected ToolResult");
+    }
+
+    // tool_result with no content
+    let json2 = r#"{"type":"tool_result","tool_use_id":"t2"}"#;
+    let block2: ContentBlock = serde_json::from_str(json2).unwrap();
+    if let ContentBlock::ToolResult { content, is_error, .. } = block2 {
+        assert!(!is_error);
+        assert!(content.is_none());
+    } else {
+        panic!("Expected ToolResult");
+    }
+}
+
+#[test]
+fn extra_unknown_fields_are_ignored() {
+    // Messages with extra fields not in our structs should parse fine
+    let json = r#"{
+        "type": "user",
+        "uuid": "x",
+        "timestamp": "2026-01-01T00:00:00Z",
+        "sessionId": "s",
+        "unknownField": "should be ignored",
+        "anotherExtra": 42,
+        "message": {"role": "user", "content": "hello"}
+    }"#;
+    let msg: RawMessage = serde_json::from_str(json).unwrap();
+    assert!(matches!(msg, RawMessage::User(_)));
+}
+
+#[test]
+fn timestamp_parsing_all_formats() {
+    // ISO 8601
+    let iso = parse_timestamp(&serde_json::json!("2026-03-22T17:00:13.192Z")).unwrap();
+    assert_eq!(iso.year(), 2026);
+
+    // Unix millis
+    let ms = parse_timestamp(&serde_json::json!(1769461914249_i64)).unwrap();
+    assert!(ms.year() >= 2025);
+
+    // Unix seconds
+    let secs = parse_timestamp(&serde_json::json!(1769461914_i64)).unwrap();
+    assert!(secs.year() >= 2025);
+}

crates/dirigent_anth/tests/usage_parse.rs

@@ -0,0 +1,101 @@
+use dirigent_anth::anth_usage::process_usage_screen;
+
+const SAMPLE: &str = r#"
+────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
+   Status   Config   Usage   Stats
+
+  Session
+  Total cost:            $0.0000
+  Total duration (API):  0s
+  Total duration (wall): 4s
+  Total code changes:    0 lines added, 0 lines removed
+  Usage:                 0 input, 0 output, 0 cache read, 0 cache write
+
+  Current session
+  ███████                                            14% used
+  Resets 12:30pm (Europe/Vienna)
+
+  Current week (all models)
+  ██████                                             12% used
+  Resets May 12, 9am (Europe/Vienna)
+
+  Current week (Sonnet only)
+                                                     0% used
+  Resets May 12, 9am (Europe/Vienna)
+
+  What's contributing to your limits usage?
+  Approximate, based on local sessions on this machine — does not include other devices or claude.ai
+
+  Last 24h · these are independent characteristics of your usage, not a breakdown
+
+  97% of your usage came from subagent-heavy sessions
+   Each subagent runs its own requests. Be deliberate about spawning them — and
+   consider configuring a cheaper model for simpler subagents.
+
+  16% of your usage was at >150k context
+   Longer sessions are more expensive even when cached. /compact mid-task, /clear
+   when switches to new tasks.
+
+  Subagents               % of usage
+  Explore                         3%
+  claude-code-guide               2%
+
+  d to day · w to week
+
+  Esc to cancel
+"#;
+
+#[test]
+fn parses_gauges() {
+    let result = process_usage_screen(SAMPLE);
+    assert_eq!(result.data.gauges.len(), 3);
+
+    assert_eq!(result.data.gauges[0].name, "Current session");
+    assert_eq!(result.data.gauges[0].percent_used, 14);
+    assert_eq!(
+        result.data.gauges[0].resets.as_deref(),
+        Some("12:30pm (Europe/Vienna)")
+    );
+
+    assert_eq!(result.data.gauges[1].name, "Current week (all models)");
+    assert_eq!(result.data.gauges[1].percent_used, 12);
+    assert_eq!(
+        result.data.gauges[1].resets.as_deref(),
+        Some("May 12, 9am (Europe/Vienna)")
+    );
+
+    assert_eq!(result.data.gauges[2].name, "Current week (Sonnet only)");
+    assert_eq!(result.data.gauges[2].percent_used, 0);
+
+    // resets_iso should be present for all gauges with reset info
+    assert!(result.data.gauges[0].resets_iso.is_some());
+    assert!(result.data.gauges[1].resets_iso.is_some());
+    assert!(result.data.gauges[2].resets_iso.is_some());
+
+    // Week resets should contain the right date
+    let week_iso = result.data.gauges[1].resets_iso.as_ref().unwrap();
+    assert!(week_iso.starts_with("2026-05-12") || week_iso.contains("05-12"));
+}
+
+#[test]
+fn parses_contributions() {
+    let result = process_usage_screen(SAMPLE);
+    let contrib = result.data.contributions.as_ref().unwrap();
+
+    assert_eq!(contrib.factors.len(), 2);
+    assert_eq!(contrib.factors[0].percent, 97);
+    assert!(contrib.factors[0].description.contains("subagent-heavy"));
+    assert_eq!(contrib.factors[1].percent, 16);
+
+    assert_eq!(contrib.subagents.len(), 2);
+    assert_eq!(contrib.subagents[0].name, "Explore");
+    assert_eq!(contrib.subagents[0].percent, 3);
+    assert_eq!(contrib.subagents[1].name, "claude-code-guide");
+    assert_eq!(contrib.subagents[1].percent, 2);
+}
+
+#[test]
+fn raw_screen_starts_with_rule() {
+    let result = process_usage_screen(SAMPLE);
+    assert!(result.raw_screen.starts_with('─'));
+}