| name | are-generate |
| description | Generate AI-friendly documentation for the entire codebase |
| disable-model-invocation | true |
Generate comprehensive documentation for this codebase using agents-reverse-engineer.
Run the generate command in the background and monitor progress in real time.
Steps
-
Read version: Read .claude/ARE-VERSION → store as $VERSION. Show the user: agents-reverse-engineer v$VERSION
-
Run the generate command in the background using run_in_background: true:
npx agents-reverse-engineer@$VERSION generate --backend claude $ARGUMENTS
-
Monitor progress by polling the latest progress log:
- Wait ~15 seconds (use
sleep 15 in Bash), then use Glob to find the latest .agents-reverse-engineer/progress-*.log file, and Read it (use the offset parameter to read only the last ~20 lines for long files)
- Show the user a brief progress update (e.g. "32/96 files analyzed, ~12m remaining")
- Check whether the background task has completed using
TaskOutput with block: false
- Repeat until the background task finishes
- Important: Keep polling even if no progress log exists yet (the command takes a few seconds to start writing)
-
On completion, read the full background task output and summarize:
- Number of files analyzed and any failures
- Number of directories documented
- Any inconsistency warnings from the quality report
This executes a two-phase pipeline:
-
File Analysis (concurrent): Discovers files, applies filters, then analyzes each source file via AI and writes .sum summary files with YAML frontmatter (content_hash, file_type, purpose, public_interface, dependencies, patterns).
-
Directory Aggregation (sequential): Generates AGENTS.md per directory in post-order traversal (deepest first, so child summaries feed into parents), and writes CLAUDE.md pointers.
Options:
--dry-run: Preview the plan without making AI calls
--eval: Namespace output by backend.model for side-by-side comparison (e.g., file.ts.claude.haiku.sum, AGENTS.claude.haiku.md)
--concurrency N: Control number of parallel AI calls (default: auto)
--fail-fast: Stop on first file analysis failure
--debug: Show AI prompts and backend details
--trace: Enable concurrency tracing to .agents-reverse-engineer/traces/