Super-Execute

Purpose

Agent team execution engine that takes an implementation plan and drives it through mandatory review gates at every level — task, phase, and PR. Nothing ships without passing every gate.

Announce at start: "I'm using the super-execute skill to implement this plan with enforced review gates."

When to Use

• You have a validated implementation plan (from writing-plans + validating-plans)
• You want mandatory code review after EVERY task
• You want phase-level compliance checking against the plan
• You want comprehensive PR review with forced fixes before merge

Do NOT use for:

• Plans with fewer than 3 tasks (use subagent-driven-development instead)
• Exploration or research (use research-to-plan)
• Writing the plan itself (use writing-plans)

Architecture

┌─────────────────────────────────────────────────────────┐
│                    SUPER-EXECUTE                        │
│                                                         │
│  Phase 0: Plan Validation ──── GATE: plan clean         │
│       │                                                 │
│       ▼                                                 │
│  ┌─────────────── Phase Loop ──────────────┐          │
│  │  ┌──────── Task Loop ──────────┐          │          │
│  │  │  Implement ──► Code Review  │          │          │
│  │  │       ▲            │        │          │          │
│  │  │       └── Fix ◄────┘        │          │          │
│  │  │            │                │          │          │
│  │  │       GATE: review clean    │          │          │
│  │  │            │                │          │          │
│  │  │         Commit              │          │          │
│  │  │            │                │          │          │
│  │  │       Next Task ────────────┘          │          │
│  │  └─────────────────────────────           │          │
│  │       │                                   │          │
│  │  Phase Compliance Check                   │          │
│  │       │                                   │          │
│  │  Phase Code Review                        │          │
│  │       │                                   │          │
│  │  GATE: phase clean ───► Next Phase ───────┘          │
│  └───────────────────────────────────────────           │
│       │                                                 │
│  Create PR + Push                                       │
│       │                                                 │
│  Comprehensive Review ◄──── (parallel: PR bots review)  │
│       │                                                 │
│  GATE: comprehensive clean                              │
│       │                                                 │
│  Address PR Bot Comments                                │
│       │                                                 │
│  GATE: all bot issues resolved                          │
│       │                                                 │
│  ✅ DONE                                                │
└─────────────────────────────────────────────────────────┘

Required Skills

• superpowers:requesting-code-review — code review dispatch after each task
• superpowers:finishing-a-development-branch — branch completion
• validating-plans — pre-execution plan validation
• comprehensive-review:full-review — full review before merge
• gh-address-comments — address PR bot review comments

Phase 0: Plan Validation

Before any code is written, validate the plan.

Steps:

1. Load the plan file
2. Invoke validating-plans skill to run full validation
3. If BLOCKERS or CRITICAL issues found:
   • Create TodoWrite tasks to fix the PLAN (not code)
   • Fix all plan issues
   • Re-run validation
   • Repeat until plan passes clean
4. GATE: Plan must have verdict ✅ PASS or ⚠️ PASS WITH NOTES

HARD GATE: Do NOT proceed to Phase 1 until plan validation passes.

Output: Validated plan file, ready for execution.

Phase 1–N: Execution Phases

Group plan tasks into phases. Each phase is a logical batch of related tasks (typically 3-5 tasks per phase, following the plan's own grouping if it has one).

Task Loop (within each phase)

For each task in the current phase:

Step 1: Dispatch Implementer

Dispatch a general-purpose subagent via Task tool:

Task tool:
  subagent_type: "general-purpose"
  description: "Implement task N"
  prompt: |
    You are implementing Task N of an implementation plan.
    
    PLAN CONTEXT:
    [Full plan text — do NOT make the agent read the file]
    
    YOUR TASK:
    [Full task text extracted from plan]
    
    COMPLETED SO FAR:
    [List of completed tasks and their commits]
    
    RULES:
    - Follow TDD: write failing test → run to verify fail → implement → run to verify pass
    - Follow the plan steps EXACTLY
    - Commit when tests pass
    - If blocked, ask questions — do NOT guess
    
    REQUIRED SUB-SKILL: Use superpowers:test-driven-development

Step 2: Code Review

After implementer completes, dispatch superpowers:code-reviewer subagent:

Task tool:
  subagent_type: "superpowers:code-reviewer"
  description: "Review task N implementation"
  prompt: |
    WHAT_WAS_IMPLEMENTED: [summary of what task N built]
    PLAN_OR_REQUIREMENTS: [full task text from plan]
    BASE_SHA: [SHA before task started]
    HEAD_SHA: [SHA after task committed]
    DESCRIPTION: [brief summary]

Step 3: Fix ALL Issues

This is non-negotiable.

• If reviewer returns Critical or Important issues:
  1. Dispatch a general-purpose fix subagent with the exact issues
  2. Fix subagent addresses EVERY issue (not "most" — ALL)
  3. Fix subagent commits fixes
  4. Re-dispatch code reviewer on the fix commit
  5. Repeat until reviewer returns clean (no Critical or Important issues)

HARD GATE: Do NOT proceed to next task until code review is clean.
Clean = zero Critical issues, zero Important issues.
Minor/Nit issues are noted but do not block.

Step 4: Commit Checkpoint

After clean review:

1. Ensure all changes are committed
2. Update TodoWrite — mark task complete
3. Record the commit SHA for phase tracking

Step 5: Next Task

Move to next task in the phase. Repeat Steps 1-4.

Phase Gate (after all tasks in phase complete)

After every task in the phase has passed its individual review:

Phase Compliance Check

Dispatch a compliance checking subagent:

Task tool:
  subagent_type: "general-purpose"
  description: "Phase N compliance check"
  prompt: |
    You are a compliance auditor. Your job is to verify that the implementation
    strictly adheres to the plan.
    
    IMPLEMENTATION PLAN:
    [Full plan text]
    
    PHASE BEING CHECKED: Phase N
    [Phase task list with descriptions]
    
    COMMITS IN THIS PHASE:
    [List of all commit SHAs and messages from this phase]
    
    YOUR JOB:
    1. git diff the full phase (BASE_SHA..HEAD_SHA)
    2. Compare EVERY requirement in the phase tasks against the actual code
    3. Flag any:
       - Missing requirements (plan says X, code doesn't have X)
       - Extra code not in plan (scope creep)
       - Deviations from specified approach
       - Test gaps (plan specifies tests that weren't written)
    
    OUTPUT FORMAT:
    - COMPLIANT: [list of requirements met]
    - NON-COMPLIANT: [list of gaps with specific details]
    - SCOPE CREEP: [list of additions not in plan]
    - VERDICT: ✅ COMPLIANT | ⚠️ MINOR DEVIATIONS | 🔴 NON-COMPLIANT
    
    If NON-COMPLIANT, explain exactly what needs to change.
    If deviations exist, they MUST have clear technical justification.

Phase Code Review

Dispatch superpowers:code-reviewer for the ENTIRE phase:

Task tool:
  subagent_type: "superpowers:code-reviewer"
  description: "Review phase N as a whole"
  prompt: |
    WHAT_WAS_IMPLEMENTED: [summary of all tasks in phase N]
    PLAN_OR_REQUIREMENTS: [full phase text from plan]
    BASE_SHA: [SHA before phase started]
    HEAD_SHA: [SHA after phase completed]
    DESCRIPTION: Phase N complete review - checking cross-task integration

Fix Phase Issues

If compliance check or phase review surfaces issues:

1. Dispatch fix subagent for each set of issues
2. Fix ALL issues (compliance gaps AND review findings)
3. Re-run BOTH compliance check and phase review
4. Repeat until both are clean

HARD GATE: Do NOT proceed to next phase until:
  - Compliance verdict is ✅ COMPLIANT or ⚠️ MINOR DEVIATIONS (with justification)
  - Phase code review has zero Critical/Important issues

Then and only then: proceed to next phase.

Post-Execution: PR Pipeline

After ALL phases complete and pass their gates:

Step 1: Create PR + Push

1. Invoke superpowers:finishing-a-development-branch skill
2. Create the PR with a comprehensive description summarizing:
   • What was built (per phase)
   • All compliance check results
   • All review results
   • Commit history

Step 2: Comprehensive Review (while PR bots work)

Immediately after pushing:

1. Invoke comprehensive-review:full-review skill
2. This runs in parallel with external PR review bots (CodeRabbit, etc.)
3. Fix ALL issues surfaced by comprehensive review:
   • Dispatch fix subagents for each finding
   • Re-run comprehensive review to verify fixes
   • Repeat until clean
4. Commit and push fixes

HARD GATE: ALL comprehensive review findings must be resolved.

Step 3: Address PR Bot Comments

After comprehensive review is clean AND PR bots have finished:

1. Invoke gh-address-comments skill
2. This reads all PR bot comments (CodeRabbit, etc.)
3. Fix ALL issues surfaced by PR bots:
   • Address every comment
   • Dispatch fix subagents as needed
   • Push fixes
4. Repeat until all bot comments are resolved

HARD GATE: ALL PR bot comments must be addressed.
Zero unresolved comments before declaring done.

Step 4: Done

Only after ALL of these pass:

• ✅ Every task reviewed and clean
• ✅ Every phase compliant and reviewed
• ✅ Comprehensive review clean
• ✅ All PR bot comments addressed

Then and only then is the work complete.

Gate Summary

Gate	Trigger	Blocks	Clean Means
Plan Validation	Before any code	Phase 1 start	No blockers or criticals in plan
Task Review	After each task commit	Next task	Zero Critical/Important review issues
Phase Compliance	After all phase tasks	Next phase	Plan adherence verified
Phase Review	After compliance passes	Next phase	Zero Critical/Important across phase
Comprehensive Review	After PR created	Bot comment addressing	All findings resolved
PR Bot Comments	After bots finish	Completion	All comments addressed

Agent Roles

Agent	Type	Dispatched When	Job
Implementer	general-purpose	Each task	Write code following plan + TDD
Code Reviewer	superpowers:code-reviewer	After each task, after each phase	Find issues in implementation
Fix Agent	general-purpose	When review finds issues	Fix specific issues from review
Compliance Checker	general-purpose	After each phase	Verify plan adherence
Comprehensive Reviewer	comprehensive-review:full-review	After PR created	Deep multi-dimensional review
PR Comment Handler	gh-address-comments	After bots finish	Address external review feedback

TodoWrite Integration

Create todos that mirror the execution structure:

Phase 0: Validate plan
  └─ Fix plan issues (if any)

Phase 1: [Phase Name]
  ├─ Task 1: [name] → implement → review → fix
  ├─ Task 2: [name] → implement → review → fix
  ├─ Task 3: [name] → implement → review → fix
  ├─ Phase 1 compliance check
  └─ Phase 1 code review → fix

Phase 2: [Phase Name]
  ├─ Task 4: [name] → implement → review → fix
  ├─ ...
  ├─ Phase 2 compliance check
  └─ Phase 2 code review → fix

Post-execution:
  ├─ Create PR + push
  ├─ Run comprehensive review → fix all
  ├─ Address PR bot comments → fix all
  └─ Verify complete

Mark each todo as in_progress when starting, completed when done. Only one in_progress at a time.

Red Flags

Never:

• Skip a review gate ("it's a small change")
• Proceed with unresolved Critical/Important issues
• Let an implementer also be its own reviewer
• Skip compliance check ("the task reviews were clean")
• Skip comprehensive review ("all phases passed")
• Skip PR bot comments ("comprehensive review caught everything")
• Declare done with unresolved issues at ANY level
• Start on main/master without explicit user consent
• Parallelize implementation agents (conflicts)

If blocked:

• Stop and surface the blocker
• Do NOT guess or work around
• Ask the user for guidance

If a fix introduces new issues:

• The fix goes through review too
• No exception — review loops until clean

Example Flow

You: I'm using the super-execute skill to implement this plan with enforced review gates.

── Phase 0: Plan Validation ──
[Invoke validating-plans]
Result: ⚠️ PASS WITH NOTES (2 warnings, non-blocking)
GATE PASSED ✅

── Phase 1: Core Data Models (Tasks 1-3) ──

Task 1: User model
  [Dispatch implementer] → implements + tests + commits
  [Dispatch code reviewer] → 1 Important issue: missing index
  [Dispatch fix agent] → adds index, commits
  [Re-dispatch code reviewer] → clean ✅
  Task 1 GATE PASSED ✅

Task 2: Auth service  
  [Dispatch implementer] → implements + tests + commits
  [Dispatch code reviewer] → clean ✅
  Task 2 GATE PASSED ✅

Task 3: Session handler
  [Dispatch implementer] → implements + tests + commits
  [Dispatch code reviewer] → 2 Critical issues
  [Dispatch fix agent] → fixes both, commits
  [Re-dispatch code reviewer] → 1 Important remaining
  [Dispatch fix agent] → fixes it, commits
  [Re-dispatch code reviewer] → clean ✅
  Task 3 GATE PASSED ✅

Phase 1 Compliance Check:
  [Dispatch compliance agent]
  Result: ✅ COMPLIANT — all 3 task requirements met
  
Phase 1 Code Review:
  [Dispatch code reviewer for full phase diff]
  Result: 1 Important — inconsistent error handling across models
  [Dispatch fix agent] → standardizes error handling, commits
  [Re-dispatch phase reviewer] → clean ✅
  Phase 1 GATE PASSED ✅

── Phase 2: API Endpoints (Tasks 4-6) ──
  [... same pattern ...]
  Phase 2 GATE PASSED ✅

── Post-Execution ──

[Create PR + push]
[Invoke comprehensive-review:full-review]
  Found: 3 issues (1 security, 2 performance)
  [Fix all 3, push]
  [Re-run comprehensive review] → clean ✅
  COMPREHENSIVE GATE PASSED ✅

[PR bots finished — CodeRabbit left 5 comments]
[Invoke gh-address-comments]
  [Address all 5 comments, push]
  PR BOT GATE PASSED ✅

✅ SUPER-EXECUTE COMPLETE
  - 6 tasks implemented and reviewed
  - 2 phases compliance-checked
  - Comprehensive review clean
  - All PR bot comments addressed
  - Ready to merge

Integration

Upstream (creates the plan this skill executes):

• superpowers:brainstorming → superpowers:writing-plans → validating-plans

Or use research-to-plan which combines brainstorming + writing + validation.

Downstream (finishing up):

• superpowers:finishing-a-development-branch — branch completion + PR creation
• comprehensive-review:full-review — deep review post-PR
• gh-address-comments — address PR bot feedback

During execution:

• superpowers:test-driven-development — implementer agents follow TDD
• superpowers:requesting-code-review — code review dispatch pattern
• superpowers:systematic-debugging — if implementer hits unexpected failures