ralph-loop_skill

Overview

This skill detects user requests to run iterative AI task loops and routes them to the Ralph loop executor. It extracts the task and success criteria from explicit phrases or conversational context, then constructs and invokes the /ralph command. It supports single and concurrent loops, registry tracking, and user options like custom loop IDs, timeouts, and iteration limits.

How this skill works

The skill watches for trigger patterns such as "ralph: X", "keep trying until X", "loop until X", and context-followers like "ralph it." It extracts a concrete task and infers a completion command when none is specified (for example, mapping "fix tests" to "npm test passes"). When extraction is confirmed it generates a /ralph invocation with optional flags (--completion, --max-iterations, --timeout, --loop-id). For multi-loop operation it checks a registry and enforces a MAX_CONCURRENT_LOOPS limit before starting.

When to use it

• When you want automated iterative fixes or improvements (tests, lint, types, build)
• When you expect repeated attempts until a verification command succeeds
• To run concurrent, tracked loops across multiple tasks
• When you want the system to infer suitable completion commands
• When you need consistent loop identifiers and registry tracking

Best practices

• Use explicit completion commands when precision matters (e.g., npm test, npx tsc --noEmit)
• Provide a clear task or keep the previous conversation focused so "ralph it" can resolve context
• Include --max-iterations or --timeout for expensive or risky tasks
• Use --loop-id for reproducible, user-friendly identifiers
• Rely on inferred completions for common patterns, but confirm if ambiguous

Example use cases

• User: "ralph: migrate src/ to ESM" → invokes /ralph "migrate src/ to ESM" --completion "node runs without errors"
• User: "keep fixing until the tests are green" → starts a loop to fix failing tests with completion "npm test passes"
• User: "loop until coverage >80%" → adds tests and runs coverage until report shows >80%
• Concurrent flows: start parallel loops to fix TypeScript errors in src/ and add tests in lib/ with separate loop IDs
• Abort or resume long-running refactors using ralph-abort and ralph-resume commands