Autonomous Chain Controller

---
name: autonomous-chain-controller
source: superpowers-plus
disable-model-invocation: true
triggers:
  - "build"
  - "implement"
  - "create feature"
  - "fix and ship"
  - "end to end"
  - "full workflow"
anti_triggers:
  - "build the matrix"
  - "build a table"
  - "simple one-liner"
description: "Meta-orchestrator that auto-detects required skill chain, executes with quality gates between steps, and auto-retries on failures. User says \"build X\" — this handles brainstorming through verification."
summary: "Use when: multi-step task. Skip when: single focused action."
coordination:
  group: orchestration
  order: 1
  requires: []
  enables: [brainstorming, debate, plan-and-execute, test-driven-development]
  escalates_to: [think-twice, failure-autopsy]
  internal: false
composition:
  consumes: [goal, phased-plan]
  produces: [todo-items, implementation]
  capabilities: [orchestrates-workflow, sequences-skills]
  priority: 3
---

# Autonomous Chain Controller

> **Wrong skill?** Single action -> invoke directly. Planning only -> plan-and-execute.

**Announce at start:** "I am using the **autonomous-chain-controller** to orchestrate the full workflow."

## When to Use

- User gives vague task ("build X", "implement Y", "fix and ship Z")
- Task requires 3+ skills in sequence
- Full feature development lifecycle
- Multi-step refactoring

### Example

```bash
# Example: Chain selection for "build login page"
echo "Task: build login page"
echo "Complexity: MEDIUM (5 skills)"
echo "Chain: brainstorming -> debate -> plan-and-execute -> TDD -> verify"
echo "Gate 1: brainstorming output has >=3 options? YES"
echo "Gate 2: debate scored all options? YES"
echo "Gate 3: plan has phases? YES -> continue"
```

## Chain Protocol

### Phase 1: Classify

Task: [request]
Complexity: LOW (1-2) | MEDIUM (3-5) | HIGH (6+)
Chain: [ordered skill list]

### Phase 2: Select Chain

| Task Type | Chain |
|-----------|-------|
| New feature | brainstorming -> debate -> plan-and-execute -> TDD -> review -> verify |
| Bug fix | systematic-debugging -> TDD -> review -> verify |
| Distributed incident | debug-conductor -> (investigators forked) -> evidence-adjudicator -> failure-autopsy |
| Refactor | blast-radius-check -> plan -> TDD -> review -> verify |
| Content | brainstorming -> plan -> harsh-review -> verify |
| Investigation | adversarial-search -> investigation-state -> autopsy |

### Phase 3: Execute with Gates

Between EVERY skill:

- [ ] Previous output correct?
- [ ] TODOs logged? (todo-guardian)
- [ ] No regressions? (measurement-integrity)
- [ ] Correct branch?

Gate failure: retry < 2 -> retry / retry >= 2 -> think-twice / retry >= 3 -> HALT

### Phase 4: Completion

All skills done + gates passed + verification + TODO sweep.

---

## Failure Detectors

| Detector | Signal | Action |
|----------|--------|--------|
| Premature completion | Remaining skills | Block -> resume |
| TODO amnesia | Unlogged items | Pause -> extract -> resume |
| Branch drift | Wrong branch | HALT -> fix -> resume |
| Gate cascade | 3x same failure | HALT -> autopsy -> escalate |
| Scope creep | >8 skills | Split into sub-chains |

## Anti-Patterns

| Anti-Pattern | Detection | Correction |
|--------------|-----------|------------|
| Skipped gate | No between-skill check | Add gate, re-run previous skill |
| Wrong chain selected | Output mismatches task | Re-classify, new chain |
| Premature completion | Skills remaining | Block until all complete |
| Scope creep | >8 skills in chain | Split into sub-chains |

## Failure Modes

| Failure | Detection | Recovery |
|---------|-----------|----------|
| Wrong chain | Output mismatch | Re-analyze, new chain |
| Chain too long | >8 skills | Split into 2 sub-chains |
| Missing skill | Reference missing | Skip + log TODO |
| Retry loop | Repeated failure | think-twice, then escalate |

## Companion Skills

- **plan-and-execute**: Planning within chain
- **brainstorming**: Options at chain start
- **debate**: Design selection
- **verification-before-completion**: Final gate
- **todo-guardian**: Between-gate enforcement
- **quantitative-decision-gate**: Branch-point decisions
- **failure-autopsy**: Chain failure post-mortem