- name
SKILL.md
技能文件保留在发布者的仓库中。
在终端中粘贴复制的命令。它会把这个 SKILL.md 安装到对应的智能体中。
npx skills add BlackBeltTechnology/pi-agent-dashboard --skill implement -a cursor -y在终端中粘贴复制的命令。它会把这个 SKILL.md 安装到对应的智能体中。
npx skills add BlackBeltTechnology/pi-agent-dashboard --skill implement -a claude-code -y在终端中粘贴复制的命令。它会把这个 SKILL.md 安装到对应的智能体中。
npx skills add BlackBeltTechnology/pi-agent-dashboard --skill implement -a codex -y在终端中粘贴复制的命令。它会把这个 SKILL.md 安装到对应的智能体中。
npx skills add BlackBeltTechnology/pi-agent-dashboard --skill implement -a grok -y在 Grok Bot 聊天中发送复制的消息。Bot 会从 GitHub 获取这个 SKILL.md,并添加到你所有 Bot 共享的技能中。
Install this skill: https://github.com/BlackBeltTechnology/pi-agent-dashboard/tree/HEAD/.pi/skills/implement来自发布者
Two halves: how to write the change (discipline) and how to land it in the running system (rebuild + restart).
The skill exists because both halves are easy to get wrong:
flowchart TD
A["Did you edit code?"] --> B["src/extension/"]
A --> C["src/server/ or src/shared/"]
A --> D["src/client/ (dev mode)"]
A --> E["src/client/ (prod mode)"]
A --> F["Multi-component / openspec"]
B --> B1["npm run reload"]
C --> C1["npx tsx .pi/skills/implement/scripts/restart-server.ts<br/>(no build step: Node's native strip-only TS loader runs TS directly)"]
D --> D1["nothing, Vite HMR"]
E --> E1["npm run build && restart"]
F --> F1["npx tsx .pi/skills/implement/scripts/full-rebuild.ts"]
Quick check current mode:
npx tsx .pi/skills/implement/scripts/check-mode.ts # prints "dev" or "production"
Restart server (preserves mode unless overridden):
npx tsx .pi/skills/implement/scripts/restart-server.ts # graceful restart, keeps mode
npx tsx .pi/skills/implement/scripts/restart-server.ts --dev # force dev mode
npx tsx .pi/skills/implement/scripts/restart-server.ts --prod # force production mode
Full rebuild (after openspec-apply or multi-component change):
npx tsx .pi/skills/implement/scripts/full-rebuild.ts
full-rebuild.tsdeploys the checked-out dev version to the local running instance (build + restart + reload). It is NOT a feature-implementation step — worktree / Docker-isolated feature work does not run it. The code-review gate is separate (below).Scripts are TypeScript (cross-platform). All invocations use
npx tsxso they work identically on Linux, macOS, and Windows.tsxis already a project dep.
Full matrix with edge cases (dev-mode fallback, fault-tolerant restart, single-restart-path rule) lives in references/rebuild-matrix.md.
Review is split by moment. The inner loop runs on an unlimited engine every non-trivial change; the rate-limited cloud gate is reserved for the PR.
Inner loop (during dev, before commit) — review-code discipline. After writing a non-trivial change, review the diff with the review-code skill (eng-disciplines): engine-agnostic — inspect design→correctness→complexity→tests→naming→security, emit labelled findings, fix issue(blocking) surgically, re-review until no blocking finding remains, then commit. Runs on a model engine — no cloud quota spent, so run it freely per change.
Ship gate (opt-in, PR-time) — CodeRabbit. Reserved for the pull request so its quota is unspent during dev. Worktree-safe and server-independent — no build, no restart:
RUN_CR_REVIEW=1 npx tsx .pi/skills/implement/scripts/review-changes.ts # opt in (uncommitted)
npx tsx .pi/skills/implement/scripts/review-changes.ts --ship -t committed --base main
npx tsx .pi/skills/implement/scripts/review-changes.ts # default: skips → use review-code
Warn-and-continue, never blocks: CodeRabbit is cloud rate-limited; on limit / missing CLI / auth failure it prints "deferred to a later cycle" and exits 0. Fix Critical/Warning findings, then commit. See the code-review skill for the CodeRabbit severity triage + fix loop.
openspec-apply: the
review-codeinner-loop pass runs after each task's code is written; the CodeRabbit gate is opt-in at ship (ship-change owns the PR-time review). Both run in the worktree without touching the main server.
The builder (this loop) owns all decisions and code writes — coherence stays in one context. Spawn a subagent (explicit Agent call) only at these signals, to keep the main context sharp:
Review stays a skill (review-code), not a subagent — review+fix is coherence-critical and wants full context. Tests: run+capture inline (tee→grep); root-cause via systematic-debugging inline. Full rationale: docs/skills-as-subagents.md.
The full code-discipline reference lives in references/code-discipline.md. It expands AGENTS.md "Code Instructions" with concrete patterns, anti-patterns, and examples. Headline rules:
These rules also live in AGENTS.md so they're always in context. This skill loads on implementation triggers so they get foregrounded when the agent is about to write code.
Never rerun npm test to inspect errors. Pipe once, grep many times:
npm test 2>&1 | tee /tmp/pi-test.log # run once, capture
grep -nE 'FAIL|Error|✗|✘' /tmp/pi-test.log # find failures
grep -n -A 20 'FAIL ' /tmp/pi-test.log # failure + context
For deeper test triage (per-package vitest configs, watch mode, coverage), see the debug-dashboard skill — its references/test-failure-triage.md.
Implementing != debugging. If the dashboard starts misbehaving (server hung, bridge won't connect, blank page, tests failing for non-obvious reasons), switch to the debug-dashboard skill.
If your change went red in CI after git push, switch to the ci-troubleshoot skill.
openspec-new-change — capture a non-trivial change as an OpenSpec proposal firstopenspec-apply-change — implement tasks from an OpenSpec change with the artifact workflowopenspec-verify-change — validate implementation matches artifacts before archivingreview-code — inner-loop diff review discipline (unlimited engine), before commitcode-review — the opt-in CodeRabbit ship gate (PR-time)debug-dashboard — diagnose a misbehaving running systemci-troubleshoot — diagnose failed CI runs after pushrelease-cut — when the change is ready to ship as a versioned releaseDisciplined implementation in pi-agent-dashboard: the rebuild matrix (extension→reload, server→restart, client→build+restart, openspec-apply→full rebuild) plus the project''s code discipline rules. Us
---name: implementdescription: Disciplined implementation in pi-agent-dashboard: the rebuild matrix (extension→reload, server→restart…---# .pi/skills/implement/SKILL.md概览
Disciplined implementation in pi-agent-dashboard: the rebuild matrix (extension→reload, server→restart, client→build+restart, openspec-apply→full rebuild) plus the project''s code discipline rules. Use when writing code, after editing extension/server/client, after applying an OpenSpec change, when unsure what to rebuild, or before committing. Triggers: "rebuild and restart", "TDD this".
常见问题
从发布者处下载 .pi/skills/implement/SKILL.md,或复制路径后打开仓库。PlugMCP 只收录该技能,不托管文件。
发布者将它存放在 https://github.com/BlackBeltTechnology/pi-agent-dashboard。
Claude、Cursor、Codex 和 Grok Build 可以读取你从发布者仓库添加的 SKILL.md。在聊天中发送技能的 GitHub 文件夹链接,Grok Bot 就会安装它。