菜单
Offload dispatch to cheap/fast omp models via pi-agent.sh — name-addressed sub-agent verbs (start/send/poll/peek/ls/stop/watch) over background omp workers with idempotent poll, worktree isolation, and distilled reports. Main loads this to write the offload usage it embeds in a builder brief.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | pi-dispatch |
| description | Offload dispatch to cheap/fast omp models via pi-agent.sh — name-addressed sub-agent verbs (start/send/poll/peek/ls/stop/watch) over background omp workers with idempotent poll, worktree isolation, and distilled reports. Main loads this to write the offload usage it embeds in a builder brief. |
scripts/pi-agent.sh is the name-addressed unified entry point over the omp
worker primitives. The registry is the filesystem: $PI_RUNS_DIR/agents/<NAME>
symlinks to the run's RUNDIR. Main (the orchestrator) loads this usage and
embeds it verbatim into a builder brief when offloading; the builder operates
the verbs as a pure operator.
| verb | command | purpose |
|---|---|---|
| dispatch a worker | pi-agent.sh start NAME [--acp] [--profile P] [BRIEF] | launch a batch (or interactive ACP) worker in the background |
| follow-up turn | pi-agent.sh send NAME TEXT_OR_FILE | resume a finished worker's session with context (SendMessage semantics); on an ACP session, start the next turn |
| status poll | pi-agent.sh poll NAME | one-shot one-line status: RUNNING or a terminal STATUS=OK|FAIL … |
| activity snapshot | pi-agent.sh peek NAME | one-shot agent-view snapshot of a live run |
| agent panel | pi-agent.sh ls | list registered agents + their state |
| cancel | pi-agent.sh stop NAME | idempotent group-kill + unregister |
| background notifications | pi-agent.sh watch [INTERVAL] | BLOCKING; polls every registered agent, prints one line per meaningful state change, exits when nothing is in flight |
pi-worktree.sh create and put its ABSOLUTE path in the brief.pi-agent.sh start per task and pi-agent.sh watch as
its main loop; relay PERMISSION lines back to main, and run each
worker's acceptance check when it settles.pi.stream.jsonl, distilled final text as result.md.profiles.conf maps profile names to omp models. Precedence:
PI_PROVIDER/PI_MODEL env > --profile/PI_PROFILE > default
(grok-build). Pick fast for mechanical work, balanced for ordinary
implementation, careful for harder reasoning; the reviewer's profile must
be ≥ the builder's.
omp installed and authenticated (omp → /login), jq, git (for
worktrees). Probe with pi-probe.sh before first dispatch in a session.
Interactively create `.obsidian.yaml` and install starter templates (task / doc / adr) for this project. Triggers via `/obw:init` or when another obw skill reports missing config.
Unified entry point for Obsidian daily-note captures and long-form notes. Triggers on "記一下 / log / 紀錄 / capture this / 寫到 journal" (→ cap mode) and "建立筆記 / new note / 寫一份筆記 / create a note on" (→ note mode). Also via `/obw:jot`. Requires `.obsidian.yaml`.
Obsidian Workspace PM — tasks, documents, and ADRs in your Obsidian vault. Triggers on task/doc/ADR lifecycle requests ("add a task", "create an ADR", "archive X", "list in-progress tasks", "refresh dashboard") and via `/obw:pm`. Requires `.obsidian.yaml` with a `pm.project` section.
Render markdown or Mermaid content as formatted HTML in the browser. Triggers when the user asks to view, render, or preview a document as HTML; when the user asks to visualize, diagram, chart, or draw architecture, flows, sequence/class/state/ER diagrams; when resolving plan files from ~/.claude/plans/; or proactively when about to output a table with 4+ rows or 3+ columns, a structured comparison, an audit, a feature matrix, or any formatted content exceeding ~50 lines in the terminal.
Use for live browser automation via the agent-browser CLI — open URLs, snapshot pages, click/fill/screenshot, inspect elements, scrape or read web pages. Ref-based workflow for AI agents. Not for writing Playwright test files (use playwright or web-test).
Use to write or set up Playwright / @playwright/test E2E tests — test specs, locators, assertions, fixtures, config. Focuses on structuring tests with high-precision locator strategies. Not for live page exploration or debugging (use agent-browser or web-test).