Use when spawning a task()/agent()/workpool() subagent with an outputSchema, to reduce yield-tool retry failures by stating the exact result envelope and required fields in the task prompt.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
설치 방법 보기설치하기 전에 에이전트에게 전달되는 지시 내용을 확인할 수 있습니다.
When a subagent is spawned with an outputSchema, it must finish by calling the hidden yield tool with a result matching that schema. On a cold turn, models frequently get the wrapper or field names wrong on the first attempt:
{"data": <schema-matching object>} on success or {"error": "<message>"} on failure — not the bare object.title/body/priority/confidence/file_path/line_start/line_end, or a research schema needing summary/architecture/files)."result" as an incremental-yield section label instead of one of the schema's own top-level property names.OMP's built-in retry loop (MAX_YIELD_RETRIES=3) recovers from this ~99% of the time, so it rarely loses work — but each failure burns one extra subagent turn (tokens + latency) that a clearer prompt avoids.
When calling task(), agent(), or workpool() with an outputSchema, append an explicit output-contract line to the task/prompt text (not just relying on the schema being attached mechanically):
Output contract: finish by calling yield with {"data": {...}} on success (or {"error": "<message>"} if you cannot complete the task). The data object must include these required fields verbatim: <list the schema's required top-level field names>. Do not wrap the object in an extra "result" key.
For schemas with nested/array sections (e.g. a findings[] array of review items), spell out the nested required fields too, since these are the most common source of mismatch.
This does not require any OMP config change — it is a parent-side prompting habit, applied every time an outputSchema is supplied to a spawned subagent.
Do not attempt to "fix" this via ~/.omp/agent/config.yml or hooks — the yield envelope format and retry mechanism are implemented in upstream oh-my-pi's task executor (packages/coding-agent/src/task/executor.ts), not in user config. There is no dotfiles-side lever.
아직 리뷰가 없습니다. 사용해 본 소감을 남겨 주세요.
로그인 없이 쓸 수 있습니다. 사용 후기, 질문, 작동하지 않은 환경 등을 편하게 남겨 주세요.
아직 댓글이 없습니다. 첫 댓글을 남겨 보세요.
Quick-save the last assistant response as a verbatim note in the 2ndBrain Obsidian vault (~/2ndBrain/quick-notes/<year>/). Use when the user asks to quick-save/save the last response to 2ndBrain, or invokes $2ndbrain or /skill:2ndbrain.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Write commit messages under the agent-commits convention - a Conventional Commits fork with no chore catch-all, intent-only type tokens, facts in trailers, and legally-grounded AI provenance (Assisted-by, never Co-authored-by). Use when composing a commit in a repo whose commitlint.config.js extends agent-commits, when deciding between feat/fix/refactor/perf, or when recording AI participation.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Coordinate multi-agent work through durable artifacts in git instead of chat. Use this whenever a task involves sending a message to another agent, agent mailboxes or handoffs, claiming work from a shared queue, notifying another agent or crew (OpenClaw, Grok Bot, Claude Code, Codex, Gas Town/Gas City workers) that work is ready, or whenever the user mentions the artifact bus, message bus, agent mail, beads mail, doorbells, or pings between agents — even if they don't say "bus" explicitly.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Use aven to find tasks, update status, and leave durable handoff context.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Plan, order and deliver a batch of backlog items (e.g. every High item) in a Conventional Docs repo, one decision, branch, worktree, draft PR, review and merge per item. Use when asked to work through the backlog, implement all items of a priority, or sequence several backlog items end to end.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Codex workflow tips from Boris Cherny (creator of Codex) and the Codex team. Use when: setting up Codex, optimizing workflows, running parallel sessions, configuring AGENTS.md, using skills/commands, subagents, hooks, MCP integrations, or learning best practices. Covers: git worktrees, plan mode, verification, permissions, Slack MCP, BigQuery, prompting tips, plugins, custom agents, sandboxing, keybindings, status lines, output styles, customization, /simplify for code quality, /batch for parallel code migrations, /loop for scheduled tasks, code review agents, /btw for mid-task questions, /effort max reasoning, remote control sessions, voice mode, setup scripts, session naming, /color, PostCompact hook, auto mode, /schedule cloud jobs, iMessage plugin, auto-memory, and auto-dream, mobile app, session teleporting, Cowork Dispatch, Chrome extension, Desktop app, /branch forking, --bare SDK startup, --add-dir multi-repo, --agent custom agents, /voice input.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.