| name | are-plan |
| description | Compare AI planning quality with and without ARE documentation (experimental) |
| disable-model-invocation | true |
Compare AI planning quality with and without ARE documentation.
Run the plan command to create an A/B comparison of AI planning quality.
Steps
-
Read version: Read .claude/ARE-VERSION → store as $VERSION. Show the user: agents-reverse-engineer v$VERSION
-
Run the plan command:
npx agents-reverse-engineer@$VERSION plan --backend claude $ARGUMENTS
-
Wait for completion — the command runs two sequential AI planning sessions (without docs, then with docs) and outputs a comparison table.
-
On completion, summarize the comparison results:
- Quantitative metrics (file references, actionable steps, code identifiers)
- Cost and latency comparison
- Quality evaluation scores (if
--eval was used)
This creates two git branches from HEAD and runs an AI planner in each:
- With ARE docs: Full documentation available
- Without ARE docs: Source code only (ARE artifacts stripped)
The comparison measures how much ARE documentation improves AI planning quality.
Options:
--eval: Run AI quality evaluator on both plans (5-point rubric)
--eval-model <name>: Model for the evaluator (default: same as --model)
--model <name>: AI model to use for planning
--dry-run: Show what would happen without executing
--list: List all saved plan comparisons
--show <id>: View a previous comparison by date
--force: Overwrite existing branches for the same task
--debug: Show verbose AI execution details