standardbeagle-tools

Record + replay an interactive guided demo overlay on the live app — self-advancing walkthrough, labelled gesture affordances. 瀏覽器互動導覽。 Use when: demo a feature, create walkthrough, run walkthrough tool, guided product tour, replay a flow, show what was built, onboarding tour, step-by-step demo overlay.

Adversarial cooperation loop for code quality verification with plan adjustment at each phase. 對抗協作循環：實施者與驗證者互相制衡，逐階調整計劃，確保代碼品質。 Use when: execute dart task, run quality pipeline, adversarial review, implement with verification

Search + synthesize prior agent session history. Claude Code primary (~/.claude/projects/); Codex/Cursor secondary. Discover -> branch/keyword filter -> extract skeleton/errors -> session-historian subagent. Use when: what was tried, prior attempts, past decisions, recent work, why approach abandoned, cross-session context, even without saying 'sessions'.

Canonical verdict-only return schema for adversarial reviewer subagents. Reviewers emit compact ≤30-line verdict block; main loop parses verdict + blockers only. Use when: authoring or updating reviewer agent, building verdict consumer, debugging reviewer report parsing.

Diagnose + configure compound-engineering env — CLI deps, plugin version, repo-local config. Guided install for missing tools, gitignore + config bootstrap. Use when: troubleshoot missing tools, verify CE setup, onboard repo, refresh config template, /ce-setup re-check. Skip: unrelated to compound-engineering tooling.

Resolve review feedback by evaluating validity and fixing issues in parallel. Pre-commit-first. Use when: applying review feedback from local reviewer agent output, resolving pre-commit review findings, working through code-review agent suggestions, addressing reviewer-flagged issues before commit. Also: PR comments via gh as secondary path. 解決審查回饋：本地審查代理輸出優先，PR次之。

Multi-source research recipe via caveman-mcp — gather URLs/PDFs/PRs in parallel, compress each, synthesize with provenance. Pairs with knowledge-hygiene for conflict detection. 多源研究：並行採集、壓縮、含出處合成。 Use when: research topic across sources, prep literature review, ingest reading list, build evidence pack, multi-source synthesis with citations

"Start Ralph Wiggum adversarial cooperation loop on dartboard with plan adjustment. 在看板上啟動Ralph Wiggum對抗合作循環，含計劃調整。 Use when: start execution loop, run dartboard tasks, begin adversarial loop, automate task execution, process task queue"

Interrogate any new task before it is persisted — extract intent, flow, domain terms, design patterns, scope, and verification with tier-gated depth. 任務創建時審問之，層層深掘以求精確。 Use when: plugin creates task, schedule work, grill request, interrogate task before planning, task intake, dartai task creation

Architecture/security/performance claims need >=2 independent sources or escalate. Load-bearing = wrong cascades downstream. Use when: spec/ADR architectural decisions, security primitive choice, performance characteristic assertion, PR with load-bearing claim, subagent design doc. Skip: typos/comments, single authoritative source documented, exploratory hypothesis.

New memory entries need ISO 8601 timestamp + source-event tag (loop-id/task-id/commit-sha) for temporal conflict resolution. Soft guidance — backfill on next touch. Use when: writing new memory, creating per-memory file, reviewing PR adding memory, subagent appending memory. Skip: existing files lacking fields, CLAUDE.md preferences.

生成、批判、呈現並收斂構想。Generate, critically evaluate, present, and converge ideas into ranked options, principles, decision trees, or implementation-plan artifacts. Use when: 'what should I improve', 'give me ideas', 'ideate on X', 'surprise me', 'what would you change', want options or a visual HTML idea report. Skip: clear bug with repro, mechanical refactor, direct implementation with complete spec.

Grounded research pipeline: gather >=2 sources -> conflict-detect -> synthesize-with-provenance, multiview retrieval. 多源研究管道。 Use when: synthesizing claims from multiple docs/repos/papers, before load-bearing assertion in spec/design, fact-checking PR, gathering evidence for architectural decision, validating contradicting claim. Skip: single authoritative source suffices.

Detect rationalization trap long chain-of-thought before small load-bearing change, esp. CoT defending prior decision against fresh evidence. 理性化陷阱檢測。 Use when: long CoT before small commit, PR whose Why dwarfs diff, ideation picks overriding high-confidence bullets, post-hoc audit of decision feels off. Skip: exploratory CoT with no commit, change big enough that long reasoning is proportionate.

Run multi-source-research + rationalization-trap-check on each load-bearing claim in target doc, spec, or PR. Surfaces conflicts + rationalization-traps before commit/merge. 對目標跑多源+陷阱檢核。 Use when: pre-merge fact-check on doc/PR, post-spec audit before locking architectural decision, validating claim contradicting Phase 0 high-confidence bullet, periodic audit of /docs/research/ outputs.

Generate a self-contained Tailwind+Mermaid HTML report from structured data, write to OS temp dir, open in browser. Reusable presenter for audits/reviews/reports. Use when: present an HTML report, render audit/review results, show a dashboard, visualize findings, before hand-rolling report HTML

Open the public core mini-IDE/editor for interactive review: cards, strategy cards, summary confirmation, demos, decisions, and annotate-artifact screens. Use when: a skill or agent needs browser-based editing, card sorting, decision review, artifact annotation, layout-gated demos, or HITL feedback loops. Skip: read-only report output → present:html-report; reopening an existing doc → present:doc

Use when starting any conversation — establishes how to find and use skills, requiring skill invocation before ANY response including clarifying questions. 啟話即用：定技能發現與調用之法，凡答（含反問）前必先察技能。

Full adversarial quality loop — implement, self-attack, parallel verification, quality gates, final validation. 全品質循環：實現、自攻、並行驗證、質量關卡、最終確認. Use when: implement task with quality, run adversarial review, verify code quality, run quality gates, complete workflow task

Stream user interactions from browser overlay in real time via agnt watch + Monitor - panel messages, clicks, sketches, design chats. 透過agnt watch + Monitor即時串流瀏覽器覆蓋層用戶互動：面板訊息、點擊、草圖、設計對話。 Use when: watch user interactions, real-time browser events, panel message stream, sketch events, design mode stream

Track user interactions and DOM mutations for debugging click handlers, form inputs, and dynamic UI changes. 追蹤用戶互動與DOM變動，除錯點擊處理器、表單輸入、動態UI變化。 Use when: debug click handler, track interactions, DOM mutations, form validation debug, dynamic content debug, mutation rate

"Set up project with dev standards — rules, skills, and hooks. 配置項目以開發規範，立規則、技能、鉤子. Use when: setup dev standards, configure project rules, install dev hooks, bootstrap project standards, apply standard rules"

Create new project-specific skill in `.claude/skills/` to codify recurring pattern. 在項目中新建可複用技能。 Use when: add skill, create skill, new skill, codify pattern, make reusable workflow

dev-standards 設定網關，導向正確之手動配置技藝。Routes project-setup / config-CRUD intent to the right manual dev-standards skill. Use when: bootstrap project dev standards, re-detect tech stack, refresh CLAUDE.md/hooks, add/edit/remove project rules, create project-specific skill, SKILL.md description style, diagnose/configure compound-engineering env, /ce-setup, unsure which config skill.

記錄已解決問題以複利機構知識。Document solved problems in docs/solutions/ with structured YAML frontmatter. Use when: non-trivial problem solved + verified, prevent re-discovery, capture institutional learning before context fades, document best practice or workflow pattern. Skip: trivial (typo), solution unverified, problem in progress, project lacks docs/solutions/ convention.

Browser element inspect, layout diag, navigation (back/forward/reload/goto) via proxy exec — target iframe-wrapped, scope target:inner/outer; use __devtool.* helpers, not raw querySelector/getComputedStyle. 瀏覽器元素檢測、佈局、導航；目標恆iframe包裹。 Use when: inspect element, debug layout, z-index, overflow, box model, flex/grid, navigate page, goto URL, reload, iframe scoping, before raw DOM JS in proxy exec

Index + router for agnt browser skills — agnt drives real browser via reverse proxy: dev server, live page navigation and inspection, DOM/CSS/layout diagnostics, console and network errors, screenshots, sketch mode, a11y/performance/security/SEO audits. Use when: picking agnt skill, debugging browser issue, inspecting element, auditing page, starting dev server or proxy.

Routes caveman / context-compression intent to right manual skill — sole auto gateway; others manual. caveman 唯一自動網關。 Use when: compress file/PDF/slides/spreadsheet/audio, condense git diff/log or PR URL, OCR/summarize image/screenshot/diagram, compress error logs/stack traces/notes, fetch+compress URL (YouTube/GitHub/arXiv), multi-source research, caveman-mcp setup, unsure which caveman skill.

Routes Dart / dart-query task-management intent to right manual skill — sole auto gateway; other dart-query skills manual. dart-query 唯一自動網關。 Use when: Dart tasks or dart-query, create/query/update/delete tasks, batch or DartQL ops, task lifecycle or relationships, watch tasks, recurring tasks, PM recipes, workspace docs/time, set up dart-query, unsure which dart skill.

Routes design-utilities intent to right manual skill among color, typography, design-token, image, a11y helpers — sole auto gateway; others manual. design-utilities 唯一自動網關。 Use when: color/contrast/palette, typography/font/type scale, design tokens, image optimize/responsive/blurhash, static a11y/WCAG audit, unsure which design-utilities skill.

Lazy-senior-dev build gate: write least code works. 最懶開發者之建造閘。 Ladder: YAGNI → stdlib → native platform → installed dep → one line. Use when: over-engineering, bloat, boilerplate, unrequested abstraction, new dep for few lines, 'be lazy', 'simplest solution', 'yagni', 'do less', before scaffolding. Skip: trust-boundary validation, data-loss handling, security, a11y, user asked full version.

聲稱完成/修復/通過前必驗證。Run verification commands + confirm output before claiming complete/fixed/passing — evidence before assertions. Use when: about to claim done/fixed/passing, before commit or PR, before delegating downstream, dartai:task-executor Phase 9 final validation. Skip: reporting in-progress with no completion claim; no verification command exists for change type.

4-layer shape for MCP tools emitting source-linked claims — source-exists, claim-derivable, no-over-extension, no-hallucinated-cite. Use when: MCP tool returns sourced facts, summaries, citations; readers follow output links; reviewing research/synthesis/doc-gen server citing URLs/file-paths/memory-IDs. Skip: pure retrieval sans claims, structural tools (list/count), output asserts nothing.

Response schema for MCP tools consuming 2+ sources — surface disagreement structurally, no silent winner-pick. conflicts[] only when sources disagree. Use when: MCP tool synthesizing across sources (research synthesis, multi-doc Q&A, cross-reference, multi-source memory recall), tool answering with 2+ snippets, output read by conflict-detector. Skip: single-source tools, pure aggregation.

Retrieval MCP progressive disclosure: default single mode; dense/lexical/symbolic/multiview opt-in flags. Index-once query-many. Use when: retrieval MCP over corpus with multiple index types (semantic + lexical + symbolic + graph), tool hitting recall cliff on multi-hop, consumers want opt-in expensive merges. Skip: single-corpus single-mode tools, all-1-hop queries, generative tools.

mcp-architect 唯一自動網關 — route MCP-server design/architecture intent to right manual skill. Use when: design or analyze MCP server, organize multi-tool MCP, tool schema/error/response design, token-efficient responses, progressive discovery, retrieval modes, citation or conflict response shapes, MCP examples, unsure which mcp-architect skill.

Routes MCP-testing / mcp-tester intent to right manual skill — sole auto gateway; other mcp-tester skills manual. mcp-tester 唯一自動網關。 Use when: test or debug MCP server, add server to debug proxy, view MCP JSON-RPC logs, inspect protocol traffic, hot-swap server binary, send raw JSON-RPC, check server status, validate MCP tool schema, set up mcp-tester, unsure which mcp-tester skill.

Native a11y primitives — focus-visible, focus-within, inert, prefers-reduced-motion, prefers-color-scheme, prefers-contrast, forced-colors, accessibility-tree for dialog/popover/anchor-positioning, ARIA labeling. Use when: focus styles, keyboard nav, animating UI, picking colors, modal/popover content, auditing component a11y. Skip: backend code, non-rendered content.

Baseline-tier decision tree for HTML/CSS/DOM Tier A Widely-9mo auto-adopt, Tier B Newly prompt, Tier C Limited avoid sans fallback. Use when: writing CSS/HTML, picking UI pattern, replacing JS lib, choosing layout/animation primitive, before popover/dialog/container-queries/anchor-positioning/view-transitions/scroll-driven-animations/color-mix/oklch/:has(). Skip: non-UI code.

Container queries, container query units (cqw/cqh/cqi/cqb), subgrid, :has() relational selector for responsive component-driven sizing. Use when: cards/sidebars/dashboards/responsive components, grandchild row alignment across siblings, styling parents by descendants, replacing viewport media queries with component-aware. Skip: static single-viewport layout, Baseline-Widely-3yr-only project.