cwijayasundara Auto Skill
Autonomous build loop implementing Karpathy's ratcheting pattern with GAN-style generator-evaluator separation, agent teams for parallel execution, sprint contracts for verifiable done-criteria, self-healing with failure-driven learning, and session chaining for multi-context-window builds.
Ultracode tip: Leave ultracode off here (
/effort highor lower). This loop already orchestrates its own agent teams and generator↔evaluator fan-out against sprint contracts; ultracode's auto-workflows would double-orchestrate, fight the contracts, and burn tokens. Do the divergent thinking earlier (/brownfield,/design,/spec) with ultracode on, then turn it off before running/auto.
Progressive loading (Phase 4)
This skill is an orchestrator index. Load only the section file for the step you are executing — do not read every reference up front.
| When | Read |
|---|---|
| SECTION 1: Usage, Prerequisites, and Agent Delegation | references/section-1-1-usage-prerequisites-and-agent-delegation.md |
| SECTION 2: Context Recovery (Step 1 of Every Iteration) | references/section-2-2-context-recovery-step-1-of-every-iteration.md |
| SECTION 3: Sprint Contract Negotiation (Steps 2-3) | references/section-3-3-sprint-contract-negotiation-steps-2-3.md |
| SECTION 4: Agent Team Execution (Step 4) | references/section-4-4-agent-team-execution-step-4.md |
| SECTION 4B: Cross-Group Parallelism | references/section-4B-4b-cross-group-parallelism.md |
| SECTION 5: Ratchet Gate (Step 5) | references/section-5-5-ratchet-gate-step-5.md |
| SECTION 6: PASS/FAIL Handling (Steps 6-7) | references/section-6-6-pass-fail-handling-steps-6-7.md |
| SECTION 7: App Lifecycle Management | references/section-7-7-app-lifecycle-management.md |
| SECTION 8: Architecture Amendment Detection | references/section-8-8-architecture-amendment-detection.md |
| SECTION 9: GAN Design Loop (Frontend Groups Only, Full Mode) | references/section-9-9-gan-design-loop-frontend-groups-only-full-mode.md |
| SECTION 10: Session Chaining | references/section-10-10-session-chaining.md |
| SECTION 11: Stopping Criteria | references/section-11-11-stopping-criteria.md |
| SECTION 12: Failure-Driven Learning | references/section-12-12-failure-driven-learning.md |
| SECTION 13: Gotchas | references/section-13-13-gotchas.md |
How to use
- Parse flags and confirm prerequisites from SECTION 1 (
references/section-1-1-usage-prerequisites-and-agent-delegation.md). - Each iteration: SECTION 2 → 3 → 4/(4B) → 5 → 6 → … as needed.
- On stop/budget/done: SECTION 10–12.
- Wiring contract tests scan this file and
references/*.mdas one corpus.
Load-bearing gate names (always visible here for harness integrity)
Gate 4 / pre-merge must continue to invoke: cycle-gate.js, coupling-gate.js, mutation-gate.js / mutation-smoke, regression-gate.js (with --replay, booting the app-under-test under HARNESS_TEST_REPLAY=1 so its DB/HTTP/LLM resolve to the recorded boundary-doubles and a missing fixture is a hard fail), contract-accessibility-default.js as specified in the section files. Full procedure lives in the references.