skills.sh
AvailableUniversal skill installation, including Claude Code and Codex.
npx skills add https://github.com/ulpi-io/skills-autonomous-engineering --skill converge-loopconverge-loop
Runs a measurable until-green or until-dry loop with explicit caps, anti-thrash, and honest termination.
WORKFLOW / 05/converge-loop <goal — e.g. 'make `pnpm test` pass' or 'remove all dead exports'>
- 01
Choose until-green or until-dry and pin the termination set.
- 02
Baseline the machine-readable signal.
- 03
Diagnose one unit, apply one minimal change, and rerun the signal.
- 04
Track progress and stop repeated failure signatures before they thrash.
- 05
Exit on convergence, exhaustion, stall, or escalation and enumerate the remainder.
▸ operating envelope
Use it deliberately.
The shortest useful definition of where this skill helps—and where it should stay out of the way.
Use when
- A task repeats until a machine-checkable condition holds.
- Tests, lint, audits, migrations, or coverage need bounded iteration.
- Each cycle can make and verify one small, attributable change.
Do not use when
- The task is one command or one edit.
- The stopping condition is subjective or cannot be measured.
- The work is open-ended exploration without an exit condition.
▸ output contract
What must be true when it stops.
Produces
- Declared termination set
- Starting-to-ending signal delta
- Iteration, budget, and exit-reason ledger
- Enumerated remaining units when not converged
Guarantees
- Every loop declares a done condition, hard cap, budget, and no-progress rule.
- The done condition is machine-checkable.
- Each iteration is smaller than the whole problem.
- Caps and stalls return converged:false rather than fabricated success.
- User-owned decisions stop the loop instead of being guessed.
Honest failure states
- exhausted — the hard cap or budget fired before the done condition
- stalled — maxStall no-progress rounds were reached
- escalated — progress requires an ambiguous, external, or destructive decision
▸ enforcement
Know which claims are executable.
Scripts are linked when this skill owns deterministic machinery. The remaining rules are explicit operating contracts and must not be marketed as hooks.
▸ install and invoke
One skill, three honest surfaces.
Claude Code plugin
AvailableInstall the full plugin, then invoke /converge-loop or let routing select it from context.
Codex plugin
Work in progressA work in progress on the codex-native-plugin branch — build it, start a new session, then invoke $autonomous-engineering:converge-loop.
▸ composes with