Skip to content

encero-systems/incan / .agents/skills/ralph-loop

ralph-loop

Verified

Run a high-thoroughness end-to-end implementation loop using clean worker worktrees, per-slice planning, implementation, explicit review/fix iteration, orchestrator integration, and final commit/PR drafting. Use when the user explicitly wants a Ralph/Wiggum-style loop, autonomous multi-agent execution, a delegated feature/bugfix carried from confirmed scope through review-ready completion, or RFC-driven implementation that should be reviewed, bumped to In Progress, decomposed into phases, and executed through nested sub-loops. Supports Codex subagents by default and external CLI workers such as OpenCode when explicitly requested and available.

FileKindSize
SKILL.mdskill md35.0 KB
reference.mdmarkdown2.6 KB

reference.md

View on GitHub
# Ralph Loop Reference

Use these templates to keep worker and orchestrator outputs consistent without bloating `SKILL.md`.

## Worker prompt contract

```md
You own one bounded slice of a larger Ralph-loop task.

Loop identity:
- <loop-id>

Repository and scope role:
- Repo: <owner/repo>
- Scope role: <implementation | verification>

Owned scope:
- <files / directories / modules>

Do not edit:
- <paths outside your scope>
- <forbidden repositories or consumer product paths>

Worktree:
- <absolute path>

Verification:
- <exact command>

Requirements:
1. Produce a short slice plan using `create-plan`.
2. Implement the slice end to end.
3. Run verification.
4. Run `review` on your slice.
5. Apply actionable findings that are in scope.
6. Re-run verification and `review` until no actionable in-scope items remain or you hit a real blocker.

Constraints:
- You are not alone in the repo.
- Do not revert others' work.
- Do not expand your scope.
- Do not use stale durable state from another loop identity.
- If your scope role is `verification`, run checks and collect evidence only; do not change consumer product code.
- If you are a child loop under an RFC parent loop, do not draft a PR description, do not propose a standalone PR, and do not own RFC lifecycle edits.
- If you are a child loop and need more decomposition, use `orchestrate-parallel-work` only for leaf workers in your owned scope; do not spawn another `ralph-loop`.
- Do not commit or push unless explicitly told.

Return:
## Slice plan
- Goal:
- Loop identity:
- Repo / scope role:
- Owned scope:
- Verification:
- Risks/blockers:

## Slice result
- Changed files:
- Verification:
- Review loop summary:
- Open questions:
- Summary:
```

## Orchestrator integration checklist

```md
## Integration checklist
- If this is RFC-driven work, update the RFC's Progress Checklist as phases land.
- Confirm every worker stayed within scope.
- Read every worker's changed-file list before integrating.
- Normalize terminology, API names, and doc wording across slices.
- Re-run repo-level verification after integration.
- Run `review` on the combined output.
- Apply actionable findings.
- Repeat until no actionable integrated items remain.
- Draft commit message and PR description.
```

## Stop conditions

Worker stop condition:

- no remaining actionable blocker or warning within owned scope, or
- a concrete blocker remains that cannot be solved without changing scope

Orchestrator stop condition:

- integrated output passes the required gate
- integrated review has no remaining actionable items
- remaining risk, if any, is explicitly documented
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 470kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall