trace-review

name	trace-review
description	Analyze recent execution traces to find recurring patterns, failures, and optimization opportunities.
when_to_use	Reach for this after `/trace-compile`, when investigating a session with a high error rate, or when looking for recurring patterns that haven't yet been turned into SEPL proposals. Do NOT use for compiling raw traces — use `/trace-compile` instead. Do NOT use for proposing fixes — use `/trace-evolve` instead.
disable-model-invocation	true
paths	[".claude/traces/*.jsonl"]
allowed-tools	["Read","Bash","Glob"]
logical	report enumerates recurring failures, file hotspots, and session-health trend with recommendations

Trace Review

Analyze execution traces stored in ~/.claude/traces/ to identify patterns across sessions.

Process

1. Read recent trace files: stat -c '%Y %n' ~/.claude/traces/*.jsonl 2>/dev/null | sort -rn | head -5 | cut -d' ' -f2- to find the 5 most recent sessions
2. Parse JSONL entries: Each line is a JSON object with type, timestamp, and type-specific fields
3. Analyze patterns:
   • Repeated failures: Commands that fail (exit_code != 0) repeatedly across sessions
   • File hotspots: Files modified most frequently — candidates for refactoring or testing
   • Wasted turns: Commands that produce no useful output or always error
   • Session health: Error rate per session, trend over time
4. Report findings: Actionable insights, not raw data

Execution Checklist

• List the 5 most-recent JSONL trace files under ~/.claude/traces/
• Parse each entry's type, timestamp, and type-specific fields
• Aggregate recurring failures (exit_code != 0), file hotspots, wasted turns, per-session error rate
• Emit the four-section markdown report (Recurring Failures, File Hotspots, Session Health Trend, Recommendations)

Entry Types

• bash: command, exit_code, output_preview, cwd
• file: tool (Write/Edit), file_path, cwd
• session_end: bash_commands, file_operations, errors, unique_files_modified

Output Format

## Trace Review (last N sessions)

### Recurring Failures
- [pattern]: happened N times across M sessions

### File Hotspots
- [file]: modified N times (consider: tests? refactor? stability?)

### Session Health Trend
- Avg error rate: N%
- Avg commands/session: N
- Trend: improving/stable/degrading

### Recommendations
- [actionable suggestion based on patterns]

Examples

Input: 5 most-recent JSONL files in ~/.claude/traces/ containing 23 bash entries with exit_code != 0 for npm test and 11 file entries touching src/auth/middleware.ts.

Output:

## Trace Review (last 5 sessions)

### Recurring Failures
- `npm test` exit 1: 23 occurrences across 4 sessions — same assertion failure in auth.test.ts

### File Hotspots
- src/auth/middleware.ts: modified 11 times (consider: missing test for session-timeout branch)

### Session Health Trend
- Avg error rate: 18%
- Avg commands/session: 47
- Trend: degrading

### Recommendations
- Add an assertion-level test for the session-timeout branch before further edits to middleware.ts

Input: traces show low error rate but rg and grep commands repeated 6+ times per session against the same paths with empty output.

Output:

## Trace Review (last 5 sessions)

### Recurring Failures
- (none above threshold)

### File Hotspots
- (no file modified ≥3 times)

### Session Health Trend
- Avg error rate: 3%
- Avg commands/session: 62
- Trend: stable

### Recommendations
- Wasted-turn pattern: 6+ near-duplicate rg/grep queries per session. Vary search terms or consult `codegraph_search` first.