Work Checkpoint

Present completed work to user for verification and determine next steps.

Critical: You run in main Claude context, so you CAN use AskUserQuestion directly.

When to Invoke

Invoke at the end of /do:* commands when verification is needed.

Skip checkpoint if:

•No significant work was completed
•User explicitly said "don't stop" or "keep going"

Process

Step 1: Gather completed work

Check recent git history and planning files in .agent_planning/<topic>/ to understand what was accomplished.

Extract:

•What was implemented (from git log and SPRINT-*-PLAN.md)
•Files created or modified
•Tests written or run
•Commits made
•Decisions logged

Step 3: Build verification checklist

For each significant unit of work, create a verification item:

markdown

## Work Completed

### 1. [Feature/Task Name]
**What was done**: [Brief description]
**Files changed**: [List of files]
**How to verify**:
- [ ] [Specific verification step - e.g., "Run `npm test` and check X passes"]
- [ ] [Another step - e.g., "Open http://localhost:3000 and test Y"]

### 2. [Another Feature/Task]
...

Step 4: Present work and ask for feedback

Display the verification checklist (from Step 3), then use AskUserQuestion with TWO questions:

json

{
  "questions": [
    {
      "question": "Any feedback on the work completed above? (Leave blank if all looks good)",
      "header": "Feedback",
      "options": [
        {"label": "All looks good", "description": "No issues or comments"},
        {"label": "Minor notes", "description": "Small things to mention but OK to continue"},
        {"label": "Has issues", "description": "Problems that should be addressed"}
      ],
      "multiSelect": false
    },
    {
      "question": "What would you like to do next?",
      "header": "Next",
      "options": [
        {"label": "Address feedback", "description": "Fix issues or incorporate notes before continuing"},
        {"label": "Continue work", "description": "Proceed to next planned task"},
        {"label": "Stop here", "description": "Pause and wait for further instructions"}
      ],
      "multiSelect": false
    }
  ]
}

Key UX points:

•User sees full work summary with verification steps BEFORE the questions
•"Other" freeform input is available for detailed comments on either question
•Two questions appear together, user can answer both at once
•If user has specific feedback, they use "Other" on the Feedback question

Step 5: Capture incomplete work (if stopping)

If user selects "Stop here" AND there is incomplete work from the session, capture it:

For each incomplete or partially completed item:

code

Skill("do:deferred-work-capture") with:
  title: "Incomplete: <item name>"
  description: |
    Work stopped before completion at user checkpoint.

    What was completed: <completed parts>
    What remains: <remaining parts>
    Files affected: <list of files>
    How to resume: <steps to continue>
  type: task
  priority: 2
  source_context: "work-checkpoint stop for <command>"

This ensures incomplete work is tracked for future sessions.

Step 6: Return action

Based on user response:

Response	Action
"Address feedback"	Return `"ACTION: FIX_FEEDBACK"` with collected issues
"Continue work"	Return `"ACTION: CONTINUE"`
"Stop here"	Capture incomplete work, then return `"ACTION: STOP"`
Custom input	Return `"ACTION: CUSTOM"` with user's input

Output Format

code

CHECKPOINT_RESULT: <ACTION>
WORK_ITEMS_REVIEWED: <n>
FEEDBACK_SUMMARY:
- Item 1: [status] - [any notes]
- Item 2: [status] - [any notes]
USER_NEXT_ACTION: <what user wants to do>
ISSUES_TO_ADDRESS:
- [If any feedback needs addressing]

Example Flow

code

User runs: /do:it implement user authentication

[... agents complete work ...]

work-checkpoint invoked, displays:

═══════════════════════════════════════════════════════════════
Work Completed - Please Verify
═══════════════════════════════════════════════════════════════

1. User Authentication Module
   What was done: Added login/logout endpoints, JWT tokens
   Files: src/auth.py, src/routes/auth.py, tests/test_auth.py
   How to verify:
   - Run `pytest tests/test_auth.py -v`
   - Test POST /login with valid credentials

2. Password Hashing
   What was done: Added bcrypt password hashing
   Files: src/auth.py
   How to verify:
   - Check passwords are not stored in plaintext

═══════════════════════════════════════════════════════════════

[AskUserQuestion with 2 questions]:
  Q1: "Any feedback on the work completed above?"
      [All looks good] [Minor notes] [Has issues] [Other: ___]

  Q2: "What would you like to do next?"
      [Address feedback] [Continue work] [Stop here] [Other: ___]

User answers:
  Q1: Other → "Need to add rate limiting to auth endpoints"
  Q2: "Address feedback"

Returns:
CHECKPOINT_RESULT: FIX_FEEDBACK
WORK_ITEMS_REVIEWED: 2
USER_FEEDBACK: "Need to add rate limiting to auth endpoints"
USER_NEXT_ACTION: Address feedback
ISSUES_TO_ADDRESS:
- Add rate limiting to authentication endpoints

Compact Mode

For small changes (1-2 items, simple modifications), use compact mode:

Single question combining verification and next steps:

json

{
  "questions": [
    {
      "question": "Completed: [brief summary]. How does it look and what's next?",
      "header": "Review",
      "options": [
        {"label": "Good, continue", "description": "Looks correct, proceed to next task"},
        {"label": "Good, stop", "description": "Looks correct, wait for instructions"},
        {"label": "Needs changes", "description": "Has issues to address first"}
      ],
      "multiSelect": false
    }
  ]
}

References

•references/verification-templates.md - Templates for different work types
•references/feedback-handling.md - How to process and act on feedback