pulse-agent

Select and run proportionate validation for changes to apps/canvas-workspace. Use when iterating on Canvas Workspace code, finishing a feature or fix, preparing release evidence, or deciding whether a full performance report is necessary.

Deep-research a topic and build a structured Pulse Canvas workspace with approved research depth, source-backed findings, progressive or final canvas creation, spatially organized frames, content nodes, and connections. Use when the user asks to bootstrap, generate, research, organize, or build an AI-created canvas.

Extend an existing Pulse Canvas frame with local, source-backed research while preserving the surrounding canvas. Use when the user asks to research, enrich, expand, verify, update, or add sources/details inside a selected or named frame, rather than bootstrapping a whole new canvas.

Operate Pulse Canvas workspaces — read user-curated context, write results, create nodes

Add or change a Pulse Canvas capability that must be available through structured agent tools and the pulse-canvas CLI. Use for cross-workspace Capability Runtime changes spanning apps/canvas-workspace, packages/canvas-cli, or packages/cli, including new live read, operate, or unsafe application actions.

Run the canvas-workspace performance evaluation, publish the latest static dashboard through the local nginx/Cloudflare Tunnel route, capture a screenshot, and summarize the report. Use when the user asks to run Pulse Canvas/canvas-workspace perf checks, deploy the performance dashboard, refresh https://jasperhu.art/apps/canvas-perf/, or send a dashboard screenshot from Feishu/remote-server.

Use when adding a new canvas node capability to Pulse Canvas (apps/canvas-workspace). Plugin nodes are the default path; host types are the documented exception. Covers both paths' exact touch points and the landmines (union/factory desync, dispatch fallthrough, agent-tool contract surface).

Use when adding a slash command to remote-server's chat-command router. Covers the ~6 parallel registries that must stay in sync, the COMMANDS_ALLOWED_WHILE_RUNNING active-run-race decision, and the describe-remote-server passthrough parity check.

Use when adding a route under /internal/* in remote-server. Covers the split auth model (global loopback middleware + per-handler bearer check that is easy to omit) and the active-run-guard decision for any route that runs an agent turn.

Use when adding a chat platform (like feishu/discord/telegram/web) to remote-server. Covers the four PlatformAdapter methods, the eager-instantiation import trap, the separate mount step, the verifyRequest security gate, and the per-adapter blocks (clarification-consume, cancel-token) that are hand-copied and easy to omit.

Use when a maintainer asks to visualize, map, inspect, explain, or compare the repository harness for root, a package, or an app as an interactive HTML reading graph.

Use when adding a shared task-metadata key to the agent-teams runtime (like round/scope/verify). This is the highest-synchrony surface in the package — one key touches the const, a typed reader, the public export, two docs, and every host that reads it. Miss one and a host breaks silently.

Use when adding a value to a task/team/agent status union in agent-teams. The landmine is that there is NO exhaustiveness guard anywhere, so a new status silently falls through dispatch gating, readiness, scope-holding, and every status filter — and you must first know WHICH surface's state machine you are extending.

Use when adding or changing a Canvas Agent tool (canvas_* / workspace_node_* in src/main/agent/tools/). Covers the two registration surfaces, the ZodObject trap, the tool-name compatibility contract, defer_loading, and the execute-class security gate.

Use when adding a built-in main-process canvas plugin (the channel / dynamic-app / webview-page-control shape) — especially one gated by an experimental flag. Covers the registration points and the flag-timing landmine that has been copy-pasted three times instead of documented.

Use when adding or extending a capability that spans main + preload + renderer (a new IPC domain or new channels on an existing one). Covers the shared-contract placement trap, the per-file boundary allowlist, the streaming pattern, the bootstrap wire describe-canvas cannot see, and the lockstep rename rule.

Use when adding a component/hook to the blessed components/ui/ set or migrating feature code onto it. Encodes the ratchet-interaction landmines learned across four expansion rounds. Deliberately contains ZERO baseline numbers — live values live in the governance test's own comments.

Use the Canvas Workspace harness in apps/canvas-workspace to launch the real Electron app with controlled profiles, inspect the renderer through CDP, capture screenshots, operate UI controls, collect logs, and clean up sessions. Use for Canvas smoke checks, visual verification, debugging, and agent-operated app validation.

Verify the Canvas Workspace onboarding case with the local harness. Use when asked to open or test onboard/onboarding/first-run Canvas behavior, capture CDP screenshots, inspect onboarding text or layers, compare the current branch against expected onboarding content, and close the disposable session afterward.

Source-backed research workflow for Pulse Canvas bootstrapping, due diligence, technical/product/market landscape analysis, current-state research, and any canvas task where Codex must gather reliable evidence before producing an artifact. Use when canvas-bootstrap needs a bundled research protocol, or when the user asks to research reliable sources before creating or organizing a canvas.

Create, read, and modify Excalidraw board plugin nodes in Pulse Canvas.

Explicitly authorized workflow that converts a requirement into an implemented change, validation evidence, commit, push, and MR for this Pulse Coder project.

Analyze the current Pulse Coder project/repository and turn vague or concrete needs into prioritized requirements, milestones, tasks, validation plans, and execution handoff. Use when user asks “分析当前项目”, “后续需求推进”, “项目规划”, “下一步做什么”, “roadmap”, “需求拆解”, “排期”, or wants a project health check plus delivery plan.

一键把 GitHub Copilot 变成本地 Anthropic-compatible 代理，让 Claude Code 免 API Key 使用。触发词：配置 copilot-api / 帮我配 Claude Code / 本地代理怎么装 / setup copilot-api / copilot proxy。

Use when running OpenCLI commands on this remote server. Prefer public API adapters and CLI Hub passthrough on Linux; browser-backed website adapters require Chrome/Chromium, the OpenCLI Browser Bridge extension, and an interactive desktop session.

Conduct high-quality multi-round research with explicit evidence tracking, source scoring, and synthesis.

Analyze runId latency and performance from devtools data. Use when asked to investigate a specific runId's timing, identify bottlenecks across LLM/Tool/Engine Plugin, or produce a fixed bilingual (ZH+EN) diagnosis report with metrics and hypotheses.

Find sessions by topic using multi-keyword matching and return session IDs

搜索微信公众号文章技能。通过微信搜索获取文章列表，覆盖科技/AI、社会热点、财经、教育、职场等各类中文资讯；可按关键词检索并返回标题、概要、发布时间、来源公众号与链接。当用户需要查找微信公众号文章、整理参考资料或快速获取文章信息时使用此技能。

Fetch latest tweets from an X/Twitter list with the built-in twitter_list_tweets tool, then summarize key updates.

Standard git workflow for handling changes on the current branch - add, commit, push, then auto-run MR generation when needed

Automatically generate concise MR titles and descriptions based on current branch diff

Summarize sessions across a multi-day range by paging session_summary in chunks, output bilingual, and write back stable soul traits.

Twitter/X data via the 6551 API. Supports user profiles, tweet search, user tweets, follower events, deleted tweets, and KOL followers.

Analyze X/Twitter accounts, tweets, and trends through the installed Twitter MCP tools

Discover and install skills from the open agent skills ecosystem

Strict research and design workflow for general coding-agent loops with evidence-backed benchmarking, iterative user discussion, and explicit execution handoff.

Intelligent branch naming based on changes, context, and best practices

Perform comprehensive code review including code quality, security, performance, and best practices

Refactor code to improve structure, readability, and maintainability without changing behavior