open-ralph

\r \r

What this skill does\r

\r This skill runs an autonomous Ralph Wiggum coding loop using the ralph CLI with OpenCode as the agent provider.\r \r It repeatedly executes the same coding prompt until:\r \r

• The success criteria are met, OR\r
• The completion promise is printed, OR\r
• Max iterations are reached\r \r The loop is optimized for free OpenCode Zen models and includes a fallback chain in case models are rate-limited, disabled, or removed.\r \r ---\r \r

When to use\r

\r Use this skill when you want autonomous coding execution such as:\r \r

• Fixing failing tests\r
• Implementing scoped features\r
• Refactoring codebases\r
• Resolving lint/type errors\r
• Running build-fix loops\r
• Multi-iteration debugging\r \r You MUST be inside a git repository before running Ralph.\r \r ---\r \r

Free model fallback order\r

\r Always attempt models in this order:\r \r

1. opencode/kimi-k2.5-free ← Best coding performance (limited time free)\r
2. opencode/minimax-m2.1-free\r
3. opencode/glm-4.7-free\r
4. opencode/big-pickle ← Free stealth model fallback\r \r If a model fails due to availability or quota, immediately retry using the next model without changing the prompt or loop parameters.\r \r

Failure triggers for fallback\r

\r Fallback if you encounter errors like:\r \r

• model disabled\r
• model not found\r
• insufficient quota\r
• quota exceeded\r
• payment required\r
• rate limit\r
• provider unavailable\r \r ---\r \r

How to run the loop\r

\r

Attempt #1 (primary model)\r

\r Run:\r \r ralph "\x3CTASK PROMPT>\r \r Success criteria:\r \r

• \x3Clist verifiable checks>\r
• Build passes\r
• Tests pass\r \r Completion promise:\r \x3Cpromise>COMPLETE\x3C/promise>" \r --agent opencode \r --model opencode/kimi-k2.5-free \r --completion-promise "COMPLETE" \r --max-iterations 20\r \r ---\r \r

Attempt #2 (fallback)\r

\r If attempt #1 fails due to model issues, rerun with:\r \r --model opencode/minimax-m2.1-free\r \r ---\r \r

Attempt #3 (fallback)\r

\r If attempt #2 fails:\r \r --model opencode/glm-4.7-free\r \r ---\r \r

Attempt #4 (final fallback)\r

\r If attempt #3 fails:\r \r --model opencode/big-pickle\r \r ---\r \r

Tasks mode (for large projects)\r

\r For multi-step execution:\r \r ralph "\x3CBIG TASK PROMPT>" \r --agent opencode \r --model opencode/kimi-k2.5-free \r --tasks \r --max-iterations 50\r \r Fallback model order still applies.\r \r ---\r \r

Plugin troubleshooting\r

\r If OpenCode plugins interfere with loop execution, rerun with:\r \r --no-plugins\r \r ---\r \r

Sanity check available Zen models\r

\r If free model availability changes, check:\r \r https://opencode.ai/zen/v1/models\r \r Update fallback order if needed.\r \r ---\r \r

Safety notes\r

\r

• Always run inside a git repo\r
• Set iteration limits to avoid runaway loops\r
• Ensure prompts contain verifiable success criteria\r
• Review diffs before merging autonomous changes\r \r ---\r \r

Example usage\r

\r Fix failing TypeScript errors:\r \r ralph "Fix all TypeScript errors in the repo.\r \r Success criteria:\r \r

• tsc passes\r
• Build succeeds\r \r Completion promise:\r \x3Cpromise>COMPLETE\x3C/promise>" \r --agent opencode \r --model opencode/kimi-k2.5-free \r --completion-promise "COMPLETE" \r --max-iterations 20\r