evaluator
Installation
SKILL.md
Evaluator Protocol
Separate the agent doing work from the agent judging it. This is more tractable than making one agent self-critical.
When to Invoke
The orchestrator calls the evaluator in these situations:
- Post-subagent, pre-merge — after implementation passes mechanical gates AND either:
- Browser-qa is available (CDT MCP connected + app running) — runtime evaluation
- Acceptance criteria require runtime testing ("user can...", "page shows...", "form validates...")
- On-demand —
evaluate <bead-id>to test an existing feature against its criteria - Post-merge —
/dm-work:post-mergeruns evaluator against closed beads
Skip evaluator when:
- XS/S tasks without acceptance criteria
- Intent review returned COVERAGE: full, DRIFT: none, GAPS: none AND no browser-qa available
- Bead has no acceptance criteria (report to orchestrator, don't run empty evaluation)
- Task has no bead (ad-hoc work)