Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Extract readable transcripts from Claude Code and Codex CLI session JSONL files
.claude/skills/aiskillstore-extract-transcripts/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | -28% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 176% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -32% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -45% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -39% | 0% |
Extracts readable markdown transcripts from Claude Code and Codex CLI session JSONL files.
bash# Extract a single session python3 ~/.claude/skills/extract-transcripts/extract_transcript.py <session.jsonl> # With tool calls and thinking blocks python3 ~/.claude/skills/extract-transcripts/extract_transcript.py <session.jsonl> --include-tools --include-thinking # Extract all sessions from a directory python3 ~/.claude/skills/extract-transcripts/extract_transcript.py <directory> --all # Output to file python3 ~/.claude/skills/extract-transcripts/extract_transcript.py <session.jsonl> -o output.md # Summary only (quick overview) python3 ~/.claude/skills/extract-transcripts/extract_transcript.py <session.jsonl> --summary # Skip empty/warmup-only sessions python3 ~/.claude/skills/extract-transcripts/extract_transcript.py <directory> --all --skip-empty
Options:
--include-tools: Include tool calls and results--include-thinking: Include Claude's thinking blocks--all: Process all .jsonl files in directory-o, --output: Output file path (default: stdout)--summary: Only output brief summary--skip-empty: Skip empty and warmup-only sessions--min-messages N: Minimum messages for --skip-empty (default: 2)bash# Extract a Codex session python3 ~/.claude/skills/extract-transcripts/extract_codex_transcript.py <session.jsonl> # Extract from Codex history file python3 ~/.claude/skills/extract-transcripts/extract_codex_transcript.py ~/.codex/history.jsonl --history
~/.claude/projects/<project-path>/<session-id>.jsonl~/.codex/sessions/<session_id>/rollout.jsonl~/.codex/history.jsonlFor querying across many sessions, use the DuckDB-based indexer:
bash# Index all sessions (incremental - only new/changed files) python3 ~/.claude/skills/extract-transcripts/transcript_index.py index # Force full reindex python3 ~/.claude/skills/extract-transcripts/transcript_index.py index --full # Limit number of files to process python3 ~/.claude/skills/extract-transcripts/transcript_index.py index --limit 10 # List recent sessions python3 ~/.claude/skills/extract-transcripts/transcript_index.py recent python3 ~/.claude/skills/extract-transcripts/transcript_index.py recent --limit 20 python3 ~/.claude/skills/extract-transcripts/transcript_index.py recent --project myapp python3 ~/.claude/skills/extract-transcripts/transcript_index.py recent --since 7d # Search across sessions python3 ~/.claude/skills/extract-transcripts/transcript_index.py search "error handling" python3 ~/.claude/skills/extract-transcripts/transcript_index.py search "query" --cwd ~/myproject # Show a session transcript python3 ~/.claude/skills/extract-transcripts/transcript_index.py show <file_path> python3 ~/.claude/skills/extract-transcripts/transcript_index.py show <file_path> --summary
Requirements: DuckDB (pip install duckdb)
Database location: ~/.claude/transcript-index/sessions.duckdb
Transcripts are formatted as markdown with:
## User## Assistant| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-06 | fail→pass | 11,607 | 2,319 | -80% | 1 | 1 | 0% | 1,818 | 1,310 | -28% | 0 | 0 | — |
case-01 | fail→fail | 17,217 | 6,693 | -61% | 1 | 1 | 0% | 3,356 | 1,361 | -59% | 0 | 0 | — |
case-02 | fail→pass | 5,675 | 24,495 | +332% | 1 | 1 | 0% | 824 | 2,272 | +176% | 0 | 0 | — |
case-03 | fail→fail | 8,260 | 5,692 | -31% | 1 | 1 | 0% | 511 | 1,247 | +144% | 0 | 0 | — |
case-04 | fail→pass | 11,638 | 3,744 | -68% | 1 | 1 | 0% | 2,191 | 1,480 | -32% | 0 | 0 | — |
case-05 | fail→pass | 13,622 | 1,849 | -86% | 1 | 1 | 0% | 2,233 | 1,225 | -45% | 0 | 0 | — |
case-07 | fail→pass | 12,213 | 1,866 | -85% | 1 | 1 | 0% | 1,981 | 1,203 | -39% | 0 | 0 | — |
case-08 | fail→pass | 5,808 | 1,842 | -68% | 1 | 1 | 0% | 921 | 1,205 | +31% | 0 | 0 | — |
case-09 | fail→fail | 5,833 | 2,729 | -53% | 1 | 1 | 0% | 954 | 1,373 | +44% | 0 | 0 | — |
case-10 | fail→fail | 9,931 | 4,455 | -55% | 1 | 1 | 0% | 1,779 | 1,657 | -7% | 0 | 0 | — |
case-20 | pass→pass | 11,727 | 3,739 | -68% | 1 | 1 | 0% | 1,848 | 1,291 | -30% | 0 | 0 | — |
case-11 | fail→fail | 9,963 | 2,422 | -76% | 1 | 1 | 0% | 1,464 | 1,330 | -9% | 0 | 0 | — |
case-12 | fail→pass | 9,566 | 1,539 | -84% | 1 | 1 | 0% | 1,534 | 1,179 | -23% | 0 | 0 | — |
case-13 | fail→pass | 6,108 | 2,005 | -67% | 1 | 1 | 0% | 1,002 | 1,289 | +29% | 0 | 0 | — |
case-14 | pass→pass | 10,638 | 4,374 | -59% | 1 | 1 | 0% | 1,746 | 1,643 | -6% | 0 | 0 | — |
case-15 | fail→pass | 9,796 | 1,891 | -81% | 1 | 1 | 0% | 1,685 | 1,234 | -27% | 0 | 0 | — |
case-16 | fail→pass | 7,855 | 1,973 | -75% | 1 | 1 | 0% | 1,347 | 1,257 | -7% | 0 | 0 | — |
case-17 | fail→pass | 8,429 | 3,543 | -58% | 1 | 1 | 0% | 1,249 | 1,187 | -5% | 0 | 0 | — |
case-18 | pass→pass | 12,714 | 1,784 | -86% | 1 | 1 | 0% | 1,943 | 1,184 | -39% | 0 | 0 | — |
case-19 | fail→pass | 13,037 | 2,633 | -80% | 1 | 1 | 0% | 2,098 | 1,238 | -41% | 0 | 0 | — |
case-21 | fail→pass | 9,150 | 5,218 | -43% | 1 | 1 | 0% | 1,539 | 1,766 | +15% | 0 | 0 | — |
case-22 | fail→pass | 14,822 | 3,358 | -77% | 1 | 1 | 0% | 2,286 | 1,457 | -36% | 0 | 0 | — |
case-23 | fail→pass | 8,592 | 5,549 | -35% | 1 | 1 | 0% | 1,564 | 2,028 | +30% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted, and 21 counted toward the lift figure. The other 2 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +65 percentage points is the difference between those two pass rates over the 21 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.