메뉴
Self-improvement mode — analyze past sessions for recurring failures, patch own code, test, commit, rebuild. Use when "evolve", "self-improve", "fix yourself", "analyze sessions", "--evolve". Do NOT use for normal coding (use /bighead) or delegating tasks (use /delegate).
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Research codebase and create spec + phased implementation plan with file-level tasks
Self-improving optimization via Karpathy autoresearch pattern. Generates → evaluates → scores → mutates prompts/descriptions in a loop. Targets — tool-selection, system-prompt, skill, decision-parser. Use when "optimize tools", "autoresearch", "improve skill X", "self-improve prompts", "optimize tool descriptions".
End-to-end project improvement — plan, branch, delegate execution, verify. Calls /plan for analysis, then delegates fixes to agents. Use when "improve project", "swarm improve", "send agents to fix", "refactor project X", "upgrade project". Do NOT use for planning only (use /plan), single tasks (use /delegate), or own codebase (use /self-evolve).
Autonomous task loop — iterate on a task until done, commit after each step, run tests. Use when "run in loop", "autonomous mode", "keep working", "don't stop", "bighead mode", "--loop". Do NOT use for one-shot tasks (just run normally) or self-improvement (use /self-evolve).
Discover and install agent skills from skills.sh (60K+ skills). Use when "find a skill", "search skills", "is there a skill for", "install skill", "skill catalog". Do NOT use for auditing existing skills (use /skill-audit) or browsing local skills (use `skills list`).
Orchestrate multiple CLI agents in parallel — delegate tasks to claude/gemini/codex/opencode/rust-code via .tasks/ system. Use when "delegate to", "spawn agents", "parallel analysis", "send task to claude", "run in background", "orchestrate agents". Do NOT use for in-process sub-agents (use spawn_agent) or single bash commands (use bash_bg).
| name | self-evolve |
| description | Self-improvement mode — analyze past sessions for recurring failures, patch own code, test, commit, rebuild. Use when "evolve", "self-improve", "fix yourself", "analyze sessions", "--evolve". Do NOT use for normal coding (use /bighead) or delegating tasks (use /delegate). |
| allowed-tools | read_file, write_file, edit_file, apply_patch, bash, search_code, git_status, git_diff, git_add, git_commit, task, finish |
| argument-hint | [focus area: patches|loops|prompts|tools] |
Analyze your own performance history and fix recurring issues. ONE improvement per iteration.
.rust-code/evolution.jsonl + recent session logsmake check, commitRESTART_AGENT if you changed agent codeReference files for diagnosis: references/diagnosis-paths.md
| Symptom | Source |
|---|---|
| Patch matching failures | crates/sgr-agent/src/app_tools/apply_patch.rs |
| Agent stuck in loops | crates/sgr-agent/src/loop_detect.rs |
| Prompt confusion | crates/rc-cli/src/agent.rs (SGR_SYSTEM_PROMPT) |
| Tool argument errors | crates/rc-cli/src/agent.rs (execute_action) |
| Schema issues | crates/sgr-agent/src/schema.rs |
finish: "RESTART_AGENT — [what you fixed]"finish: "<solo:done/>"finish: "<solo:done/> — need human input".tasks/ for the next session to pick up.make check can take 2+ minutes — don't call it in a tight loop. Run once after your change, fix issues, run once more to verify.evolution.jsonl. The loop engine reads it for scoring trends.