dev-delegate

Main Chat                          Task Agent
─────────────────────────────────────────────────────
dev-implement (orchestrates)
  → dev-ralph-loop (per-task loops)
    → dev-delegate (this skill)
      → spawns Task agent ──────→ follows dev-tdd
                                  uses dev-test tools

For each task:
    1. Dispatch implementer subagent
       - If questions → answer, re-dispatch
       - Implements, tests, commits
    2. Dispatch spec reviewer subagent
       - If issues → implementer fixes → re-review
    3. Dispatch quality reviewer subagent
       - If issues → implementer fixes → re-review
    4. Mark task complete

Task(subagent_type="workflows:dev-implementer", prompt="""
# TASK

Implement: [TASK NAME]

## EXPECTED OUTCOME

You will have successfully completed this task when:
- [ ] [Success criterion 1]
- [ ] [Success criterion 2]
- [ ] Tests pass (TDD enforced)
- [ ] No regressions in existing tests

## REQUIRED SKILLS

This task requires:
- [Language/framework]: [Why]
- Testing: TDD (test-first mandatory)
- [Other skills as needed]

## REQUIRED TOOLS

You will need:
- Read: Examine existing code
- Write: Create new files
- Edit: Modify existing files
- Bash: Run tests and verify

**Tools denied:** None (full implementation access)

## MUST DO

- [ ] Write test FIRST (TDD RED-GREEN-REFACTOR)
- [ ] Run test suite after each change
- [ ] Follow existing code patterns in [file]
- [ ] [Other non-negotiable requirements]

## MUST NOT DO

- ❌ Write code before test
- ❌ Skip test execution
- ❌ Use `any` / `@ts-ignore` / type suppression
- ❌ Commit broken code

## DEVIATION RULES

You WILL discover unplanned work. Apply these rules and track all deviations:

| Rule | Trigger | Action |
|------|---------|--------|
| **R1: Bug** | Broken behavior, errors, type errors, security vulns | Auto-fix → test → track `[Rule 1 - Bug]` |
| **R2: Missing Critical** | Missing error handling, validation, auth, logging | Auto-fix → test → track `[Rule 2 - Missing Critical]` |
| **R3: Blocking** | Missing deps, wrong types, broken imports | Auto-fix → test → track `[Rule 3 - Blocking]` |
| **R4: Architectural** | New DB table, schema change, switching libs, breaking API | **STOP → escalate to delegator** |

**Rules 1-3:** Fix automatically, test, document what you did.
**Rule 4:** Do NOT proceed. Report the architectural decision needed back to the delegator with: what you found, proposed change, why needed, impact, and alternatives.

**Unsure which rule?** Default to Rule 4 (STOP and escalate).

In your output, include: **Deviations:** N auto-fixed (R1: X, R2: Y, R3: Z), N escalated (R4: W).

## CONTEXT

### Task Description
[PASTE FULL TASK TEXT FROM PLAN.md - don't make subagent read file]

### Project Context
- Project: [brief description]
- Related files: [list from exploration]
- Test command: [from SPEC.md]

## TDD Protocol (MANDATORY)

<EXTREMELY-IMPORTANT>
**LOAD THIS SKILL FIRST:**

Before writing any code, you MUST load the TDD skill:

This loads:
- Task reframing (your job is writing tests, not features)
- The Execution Gate (6 mandatory gates before E2E testing)
- GATE 5: READ LOGS (mandatory - cannot skip)
- The Iron Law of TDD (test-first approach)

**Load dev-tdd now before proceeding.**
</EXTREMELY-IMPORTANT>

Follow the RED-GREEN-REFACTOR cycle from dev-tdd:

1. **RED**: Write a failing test FIRST
   - Run it, SEE IT FAIL
   - Document: "RED: [test] fails with [error]"

2. **GREEN**: Write MINIMAL code to pass
   - Run test, SEE IT PASS
   - Document: "GREEN: [test] passes"

3. **REFACTOR**: Clean up while staying green

**If you write code before seeing RED, you're not doing TDD. Stop and restart.**

## Testing Tools

For test options (pytest, Playwright, ydotool), load dev-test skill after dev-tdd.

Tests must EXECUTE code and VERIFY behavior. Grepping is NOT testing.

## If Unclear
Ask questions BEFORE implementing. Don't guess.

## Output
Report:
- RED: What test failed and how
- GREEN: What made it pass
- Test command and output
- Commit SHA
- Any concerns
""")

Task(subagent_type="general-purpose",
     allowed_tools=["Read", "Glob", "Grep", "Bash(read-only)"],
     prompt="""
Review spec compliance for: [TASK NAME]

## Original Requirements
[PASTE TASK TEXT FROM PLAN.md]

## Success Criteria (from SPEC.md)
[PASTE RELEVANT CRITERIA]

## CRITICAL: Do Not Trust the Report

The implementer finished suspiciously quickly. Their report may be incomplete,
inaccurate, or optimistic. You MUST verify everything independently.

**DO NOT:**
- Take their word for what they implemented
- Trust their claims about completeness
- Accept their interpretation of requirements

**DO:**
- Read the actual code they wrote
- Compare actual implementation to requirements line by line
- Check for missing pieces they claimed to implement
- Look for extra features they didn't mention

## Review Checklist
1. Does implementation meet ALL requirements?
2. Is anything MISSING from the spec?
3. Is anything EXTRA not in the spec?

## Output Format
- COMPLIANT: All requirements met, nothing extra (after verifying code yourself)
- ISSUES: List what's missing or extra with file:line references

Be strict. "Close enough" is not compliant. Verify by reading code.
""")

Task(subagent_type="general-purpose",
     allowed_tools=["Read", "Glob", "Grep", "Bash(read-only)"],
     prompt="""
Review code quality for: [TASK NAME]

## Changes to Review
Files modified: [list files]
Commit range: [BASE_SHA]..[HEAD_SHA]

## Review Focus
1. Code correctness (logic errors, edge cases)
2. Test coverage (are tests meaningful?)
3. Code style (matches project conventions?)
4. No regressions introduced

## Confidence Scoring
Rate each issue 0-100. Only report issues >= 80 confidence.

## Output Format
### Strengths
- [what's good]

### Issues (Confidence >= 80)
#### [Issue Title] (Confidence: XX)
- Location: [file:line]
- Problem: [description]
- Fix: [suggestion]

### Verdict
APPROVED or CHANGES REQUIRED
""")

Me: Implementing Task 1: Add user validation

[Dispatch implementer with full task text]

Implementer: "Should validation happen client-side or server-side?"

Me: "Server-side only, in the API layer"

[Re-dispatch implementer with answer]

Implementer:
- Added validateUser() in api/users.ts
- Tests: 5/5 passing
- Committed: abc123

[Dispatch spec reviewer]

Spec Reviewer: ISSUES
- Missing: Email format validation (spec line 12)

[Tell implementer to fix]

Implementer:
- Added email regex validation
- Tests: 6/6 passing
- Committed: def456

[Re-dispatch spec reviewer]

Spec Reviewer: COMPLIANT

[Dispatch quality reviewer with commit range]

Quality Reviewer: APPROVED
- Strengths: Good test coverage, clear naming
- No issues >= 80 confidence

[Mark Task 1 complete, move to Task 2]