mymx2/dyc

Reviews code diffs, PRs, release readiness, and project or document audits, then reports evidence-grounded findings. Use when users ask for code review, issue or PR triage, release gates, project scorecards, security/performance/coverage review, or doc proofreading. Not for debugging root causes (hunt) or drafting prose (write).

Drives a real Chrome browser through the chrome-devtools-mcp MCP server for page automation, network and console analysis, emulation, and extension testing. Use when a task needs operating a live browser: navigating, snapshotting the accessibility tree, interacting by element uid, capture a screenshot or full-page capture, console errors, traces, or heap snapshots, or fixing the MCP server setup. Not for diagnosing application logic root causes (use hunt) or judging visual quality (use ui).

插件与技能的全生命周期管理：基于 Agent Plugins Spec 创建标准插件、通过 extensions + init 模式适配多厂商（Claude Code / Qoder / Codex / CodeBuddy），以及发现、安装、编写、校验技能。Use when creating a plugin, scaffolding a multi-vendor plugin, writing plugin.json, generating init.ts, finding/installing skills, or authoring/validating SKILL.md. Not for runtime debugging of deployed skills (route to hunt) or reviewing plugin code quality (route to check).

Runs a budget-aware agent-assisted engineering health audit for instruction/config drift, hooks/MCP, verifier surfaces, and AI maintainability. Use when users ask to audit Claude, Codex, Pi, agent instructions, MCP or hooks, verifier coverage, or AI-maintainability drift. Not for debugging application code or reviewing PRs.

Finds the actual root cause of a failing test, error, crash, regression, a page rendered blank or wrong, or a request that throws, before applying fixes, proven against a known-good reference (prior version, release artifact, or "good" screenshot). Use when users report errors, crashes, broken behavior, blank rendering, failing tests, or a used-to-work state that now fails. Not for code review (route to check) or a first-impression visual complaint with no working baseline (route to ui).

Runs a six-phase research workflow that turns unfamiliar domains, real docs, source bundles, or collected material into a publish-ready comparison or reference. Use when users ask for multi-source research: deep-dives, comparing tradeoffs, compiling sources, synthesizing unfamiliar material, or turning a source bundle into a coherent reference. Not for quick lookups or single-URL fetches (route to read).

Produces PM deliverables and decision frameworks: PRD, user stories, Epic breakdown, backlog prioritization (RICE, MoSCoW), feedback analysis, competitive comparison, metrics reviews (DAU, funnel, retention), roadmaps, GTM (pricing, channels, positioning). Use when users ask to write a PRD, split stories, rank a backlog, analyze feedback or competitors, review metrics, or plan GTM. Not for engineering planning (think), open research (learn), prose polishing (write), or code review (check).

Reads URLs, PDFs, and docs, summarizing by default and returning clean Markdown when asked to convert, save notes, quote, cite, or feed downstream work; runs a web search when built-in search is thin and the user opts in. Use when users ask to read, fetch, summarize, quote, cite, convert, save notes, or save a URL, PDF, or doc, or ask for an opt-in web search fallback. Not for local text files already in the repo.

Generates a DeepWiki-style repository wiki from repository structure, config, and dependencies, producing a single REPOWIKI.md with Mermaid diagrams and source-file references. Use when the user asks for a repository report, project documentation, codebase analysis, or wiki generation. Not for answering one-off questions about a repo, code review, refactoring, general prose documentation (route to write), or source-bundle synthesis (route to learn).

Turns rough ideas into approved, decision-complete plans with validated structure before coding. Use when users ask for a design direction, a worth-doing judgment on a feature, restructuring tangled modules, or triaging a bundle of mixed requests/feedback into accept/reject buckets. Not for bug fixes or small edits.

Builds distinctive, production-grade UI: pages, components, visual polish against screenshots or first-impression complaints, typography, and DESIGN.md/design-token contracts including Tailwind export via the @google/design.md CLI. Use when users ask for UI work, design tokens, or complain a screen looks unclear, ugly, or inconsistent. Not for backend logic or runtime/behavior regressions (route to hunt); intentional design-reference comparisons stay here.

Rewrites and polishes prose in Chinese or English, removes AI-like wording, compares before/after drafts, proofreads README and CONTRIBUTING files, rewrites a getting-started section, and authors technical documentation with the writing rulebook. Use when users ask to rewrite, proofread README or CONTRIBUTING, compare draft versions, draft release notes or social copy, localize copy, write documentation, or revise coding standards. Not for commit messages or rulebook audits (use check).