verify-plan

SkillCloud & infra

Check implemented work against an executable Plan before completion or deploy by validating Specification traceability, acceptance criteria coverage, and red/green evidence

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the verify-plan skill

What this skill tells your AI

The instructions your AI receives, as published by glowingkitty/openmates in .agents/skills/verify-plan/SKILL.md and read by ahel’s review.

Instructions

You are performing a conformance check for Plan-driven work. This does not replace tests; it confirms implementation evidence matches the Plan and its linked approved Specifications.

Step 1: Read Inputs

Read the provided docs/plans/<slug>/plan.yml, linked Specifications, tasks, and the spec-driven-development guide. Review current session status and related test tracking with sessions.py status and sessions.py check-tests.

Step 2: Build The Coverage Table

Run:

python3 scripts/plan_validate.py docs/plans/<slug>/plan.yml
python3 scripts/plan_verify.py docs/plans/<slug>/plan.yml --phase complete

For every scenario and acceptance criterion, record implementation status, coverage, evidence, and risk.

Step 3: Validate Required Evidence

Pass only when:

  • Linked Specification references resolve, changed assertions have current matching fingerprints, and required surface proof is direct and current.
  • Every required acceptance criterion is covered by green evidence, a documented user confirmation, accepted waiver, or accepted blocker. Ambiguous or uncovered criteria fail.
  • Required red and green phases have evidence, and evidence subject commits match the Plan implementation state.
  • Shared product surfaces have evidence in required order: REST API/WebSocket, CLI, npm/pip SDKs, CI/daily reproduction, web, deployed Playwright visual smoke, user confirmation, then Apple.
  • Larger user-visible web/UI work has a passing or explicitly skipped V-UI-VISUAL-SMOKE review before user confirmation or completion.
  • CLI and SDK evidence uses real commands or calls against the real dev API/WebSocket path. Mocks, fixture replay, and direct function calls do not satisfy the gate.
  • Privacy/security criteria, documentation impact, assumptions, open questions, and required proof-video evidence are resolved or explicitly accepted.

Failed required checks keep the Plan active until traceable follow-up work is complete or the user accepts a waiver or blocker. Playwright green evidence is only valid after deployment to dev and execution against app.dev.openmates.org.

Step 4: Output Report

## Plan Verification

Plan: docs/plans/<slug>/plan.yml
Specifications: <approved references>
Status: pass | fail | partial

| ID | Status | Evidence | Risk |
| --- | --- | --- | --- |
| S-1 | pass | `frontend/...spec.ts` | none |
| AC-1 | pass | `python3 scripts/tests.py run ...` | none |

Gaps:
- <gap or none>

Step 5: Continue On Failure

If status is fail, resume the smallest actionable task, fix the gap, and update the durable handoff. Ask the user only for a genuinely unresolved decision.

Rules

  • Do not mark criteria complete based on intent alone.
  • Do not weaken or remove acceptance criteria to make verification pass.
  • Do not mark a Plan verified while required green evidence is missing.

Signals

GitHub stars
46
Forks
3
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
verify-plan
Source
github.com/glowingkitty/openmates