verify

name	verify
description	Use for Phase 5 of the development workflow: verification and retrospective. Runs end-to-end testing for a milestone and generates a retrospective document. Use when someone says 'verify milestone', 'phase 5', 'retrospective', 'run verification', or after all tasks in a milestone are complete.

What This Skill Does

Executes Phase 5 (Verification & Retrospective) of the AI-Assisted Development Workflow.

Prompt

INPUTS:

Read the following files:

• workflow/plan/PLAN.md
• workflow/plan/PROGRESS.md
• workflow/spec/SPEC.md (for reference)
• All files in workflow/plan/reviews/ (task review files for this milestone)
• All files in workflow/decisions/ (decision records)

Additional context:

• Git diff or changelog: {{PASTE_OR_REFERENCE — e.g., git log --oneline main..feature-branch}}
• Test results: {{PASTE_OR_REFERENCE — e.g., output of pytest}}
• Any manual testing notes: {{YOUR_OBSERVATIONS}}

ROLE:

If executing a standard milestone verification, use qa-engineer to test and code-reviewer to analyze quality. If generating the final retrospective, use senior-pm or software-architect to capture process and architecture learnings, or use the default role if not specified.

INSTRUCTIONS:

Step 1 — Verify the milestone:

1. Run the full test suite for the milestone.
2. If the project has a local deployment target, deploy locally and manually test critical user paths. Note any issues.
3. If issues are found, feed specific failures back to Phase 4 as targeted fixes — do not re-plan.

Step 2 — Review README.md:

Check that README.md is complete and accurate for the current milestone. Ensure the Overview, Getting Started, Usage, and Architecture sections reflect what was actually built. Fix any placeholders that should have been filled in during earlier phases.

Step 3 — Generate the retrospective with these sections:

1. Summary

   • What was built (1-2 sentences)
   • Milestone: {{MILESTONE_NAME}}
   • Tasks completed: {{N}} / {{TOTAL}}
   • Waves completed: {{N}}

2. What Went Well

   • Tasks that completed cleanly on first attempt
   • Spec sections that were clear and complete
   • Patterns or approaches that worked effectively
   • Decisions (from workflow/decisions/) that proved correct

3. What Needed Human Intervention

   • Tasks that failed or required multiple review cycles (reference specific task review files)
   • Ambiguous or incomplete spec sections
   • Tool/context issues encountered
   • Decisions that had to be revised

4. Spec Gaps Discovered

   • Requirements missing from SPEC.md
   • Edge cases not covered
   • Integration points that were underspecified
   • (Pull from PROGRESS.md "Spec Gaps Discovered" section and individual task review files)

5. Test Separation Effectiveness

   • Did separately-authored tests catch issues the implementation session missed?
   • Were test tasks appropriately scoped?
   • Any cases where inline testing would have been sufficient?

6. Cost & Efficiency

   • Models used per phase
   • Phases that felt over/under-invested
   • Context issues (did any task hit context limits?)
   • Role assignments — were the right roles used for each task?

7. Recommendations for Next Milestone

   • Specific changes to prompts or role definitions
   • Changes to spec detail level
   • Changes to task granularity (too big? too small?)
   • Workflow improvements
   • Decision records that should inform the next milestone's spec

8. Areas of Low Confidence

   • Sections where you had insufficient data or made assumptions
   • Metrics or counts that are approximate
   • Recommendations you're unsure about

OUTPUT FORMAT:

Markdown file for saving as workflow/retro/RETRO-{{MILESTONE_NAME}}.md. End the file with an empty ## Review Discussion section (same pattern as task review files) for human feedback.

CONSTRAINTS:

• Be specific. Bad: "Some tasks had issues." Good: "Task 2.3 failed because the API contract didn't specify error response format for 404s."
• Focus on actionable improvements, not general observations.
• Reference specific task review files and decision records by name.
• Keep it under 500 words. This should be a quick reference, not an essay.

HUMAN REVIEW PROCESS:

After you produce the retrospective, STOP and ask the human to review workflow/retro/RETRO-{{MILESTONE_NAME}}.md. Do NOT proceed or suggest next steps.

The human will review the retrospective and may add *FEEDBACK:* comments in the ## Review Discussion section.

• Respond with *AI:* comments explaining what was changed and why, then update the retrospective accordingly.
• Do not overwrite previous discussion — append new responses below existing conversation.
• The phase is complete only when the human explicitly approves. Milestone PRs may then be merged. Do NOT merge until told.