AgentSkillsCN

ralph

配置并运行由OpenCode驱动的loop.sh智能循环作业（PROMPT.md + IMPLEMENTATION_PLAN.md + ORCHESTRATOR.md）。

SKILL.md
--- frontmatter
name: ralph
description: Configure and run a loop.sh agentic loop job (PROMPT.md + IMPLEMENTATION_PLAN.md + ORCHESTRATOR.md) driven by OpenCode.
compatibility: opencode
metadata:
  short-description: Set up loop.sh iterative jobs

Ralph (Loop Job Runner)

This skill sets up a simple iterative loop job where each iteration re-reads a prompt file, does one checklist item, commits, and stops. An optional orchestrator monitors progress and adds course corrections.

Files

  • •PROMPT.md: concise instructions the worker reads every iteration (keep under ~20 lines).
  • •IMPLEMENTATION_PLAN.md: heavy reference (task order, checklist, discoveries, rules).
  • •ORCHESTRATOR.md: monitoring playbook (log review, git checks, when to intervene).

Launch

Use ./loop.sh <iterations> <prompt_file> (a generic template is included in references/loop.sh).

PROMPT.md Rules of Thumb

  • •One checklist item per iteration.
  • •First line should reference the plan file by name.
  • •Put course corrections at the top under IMPORTANT: and tell the worker to read them first every iteration.
  • •Require the worker to update IMPLEMENTATION_PLAN.md (check items off, record discoveries).
  • •If test changes were made, require running @test-quality-verifier (or follow the test-quality-verifier skill) before finishing the iteration.
  • •Output /done exactly when complete.