원클릭으로
stdout-delegate-tutorial
Minimal 3-step chained example verifying the stdout-delegate pattern works end-to-end
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
메뉴
Minimal 3-step chained example verifying the stdout-delegate pattern works end-to-end
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
| name | stdout-delegate-tutorial |
| description | Minimal 3-step chained example verifying the stdout-delegate pattern works end-to-end |
| version | 0.2.0 |
A 3-step chained Mode D demo. Each step emits one __LLM_DELEGATE__ directive that writes
its result to a tmp file; the next step reads that file to build on the previous answer.
python3 tutorial.py --step 1
Emits:
__LLM_DELEGATE__: {"prompt": "Name one obscure programming language. Reply with the name only.", "out": "tmp/tutorial/lang.txt"}
Write the LLM response into tmp/tutorial/lang.txt.
python3 tutorial.py --step 2
Reads tmp/tutorial/lang.txt, then emits:
__LLM_DELEGATE__: {"prompt": "Give one surprising fact about <lang> in one sentence.", "out": "tmp/tutorial/fact.txt"}
Write the LLM response into tmp/tutorial/fact.txt.
python3 tutorial.py --step 3
Reads tmp/tutorial/lang.txt and tmp/tutorial/fact.txt, then emits:
__LLM_DELEGATE__: {"prompt": "Given this fact about <lang>: '<fact>' — should a beginner learn it? One sentence.", "out": "tmp/tutorial/verdict.txt"}
Write the LLM response into tmp/tutorial/verdict.txt.
python3 tutorial.py --step 4
No directive — just prints the collected results:
=== Chain complete ===
Language : <lang>
Fact : <fact>
Verdict : <verdict>
If you see all three fields populated, the 3-step chain is working end-to-end.
Build or improve agent SKILL.md files by scaffolding, iterating with worktrees, running evals, registering dependencies, and proposing splits; not for general code modules, ordinary docs, or plugin packaging.
Turn a skills repo into a multi-platform plugin (Claude Code, Codex, Cursor, Gemini, OpenCode). Use when the user wants their skills installable via `/plugin install`, loadable as a Gemini extension, or discoverable via symlink on Codex/OpenCode. Also audits existing plugins for missing manifests or adapters.
Commit the current diff with a clean conventional message — the bot picks the prefix, writes the subject, and runs `git commit`. Use when the user says "bommit", "commit this", or "just commit" without dictating the message. bommit = bot + commit.