Read version: Read .claude/ARE-VERSION → store as $VERSION. Show the user: agents-reverse-engineer v$VERSION
Run the implement command:
```
npx agents-reverse-engineer@$VERSION implement --backend claude $ARGUMENTS
```
Note: When --plan-id <id> is provided, the task description is loaded from the stored plan — no need to pass it explicitly.
Wait for completion — the command runs two sequential AI implementation sessions (without docs, then with docs) and outputs a comparison table.
On completion, summarize the comparison results:
- Implementation metrics (files created/modified, lines added/deleted)
- Test results (if --run-tests was used)
- Build and lint results (if --run-build / --run-lint was used)
- Cost and latency comparison
- Quality evaluation scores (if --eval was used)

This re-uses branches created by /are-plan and executes the implementation in both worktrees:

The comparison measures how much ARE documentation improves implementation quality.

Options:

--eval: Run AI quality evaluator on both implementations
--eval-model <name>: Model for the evaluator (default: same as --model)
--model <name>: AI model to use for implementation
--task-slug <slug>: Reference existing plan by slug (default: auto-generate from task)
--plan-id <id>: Reference existing plan by ID (shown by are plan on completion)
--run-tests: Execute test suite and include results in metrics
--run-build: Execute build and verify success
--run-lint: Run linter and include error/warning counts
--dry-run: Show what would happen without executing
--list: List all saved implementation comparisons
--show <id>: View a previous comparison by date
--force: Overwrite existing branches for the same task
--debug: Show verbose AI execution details

are-implement

With ARE docs: Full documentation available during implementation
Without ARE docs: Source code only (ARE artifacts stripped)

来源信息