菜单
Spec-driven development workflow. Interviews the user, writes success criteria, generates tests, researches the codebase, produces a detailed implementation plan, and executes it with subagents.
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
Teacher mode — explains code, concepts, or architecture with clarity and depth. Use when the user says "explain", "teach me", "how does this work", "why does this", or invokes /explain.
Capture non-obvious engineering insights about the current project into a `.notes/` zettelkasten in the project root. Use this skill when the user types `/update-notes`, says "save this to project notes", "update the notes", "add that to the zettelkasten", "note this finding", "we just learned something worth saving about this project", or signals that a session uncovered an API quirk, footgun, architecture gotcha, redundancy map, consolidation candidate, convention, invariant, benchmark, or decision-with-rationale that will help future iteration on the code. Run periodically across a project's lifetime — designed to be idempotent and run many times. Do NOT use for session summaries, user preferences, task status, or TODO items — those belong elsewhere.
Generate a polished, release-quality README.md with comprehensive Playwright screenshots of every feature in a web app. Use this skill whenever the user wants a README, wants to document their app with screenshots, says "make a readme", "release readme", "document this project", "screenshot the app for a readme", or wants to prepare a repo for public release. Also use when the user asks for project documentation with visuals, or wants to showcase their app's features.
Record terminal commands as polished GIFs using VHS. Use this skill whenever the user wants to capture terminal output as a GIF, record a CLI demo, create animated screenshots of command-line tools, make terminal recordings for READMEs, or says things like "record this as a gif", "make a gif of", "capture the terminal output", "demo gif", or "screen recording of the terminal". Also use when the user is building a README and needs terminal recordings, or when they want to show off CLI tool output visually.
Tweak dotfiles configs — tmux, shell, fzf, helix, status bar, keybindings, themes, colors. Use when the user mentions dotfiles, tmux, zshrc, aliases, keybindings, status bar, fzf config, shell config, helix config, theme, colors, powerkit, or any config file modification.
Deploy any project to Cloudflare Pages with password protection. Use when the user wants to deploy a site, put something on the internet, or share a project via URL. Handles static sites, React apps, and any folder of files.
| name | create-plan |
| description | Spec-driven development workflow. Interviews the user, writes success criteria, generates tests, researches the codebase, produces a detailed implementation plan, and executes it with subagents. |
Execute the following phases in order. Do not skip phases. Confirm with the user before proceeding to the next phase.
Launch the spec-interviewer agent via the Task tool to conduct a structured interview:
Task(subagent_type="general-purpose", prompt="You are a spec-interviewer agent. [paste spec-interviewer.md instructions]. Interview the user about: $ARGUMENTS")
Or use AskUserQuestion directly to probe:
Ask 2-3 questions at a time. Do not overwhelm.
Extract declarative, testable success criteria from Phase 1. Each criterion must be:
Write the spec to .claude/specs/{feature-name}-spec.md using the template from spec-template.md in this skill's directory. Read the template first.
Present the spec to the user and confirm before continuing.
Launch the test-generator agent via the Task tool with the spec path from Phase 2:
Task(subagent_type="general-purpose", model="sonnet",
prompt="You are a test-generator agent. Read the agent instructions at maintained_global_claude/agents/test-generator.md, then execute all 7 phases. Spec file: .claude/specs/{feature-name}-spec.md")
The agent handles:
test, test-verbose, test-cov recipesjust test to verify the red phase## Test File Locations sectionPresent the test generation report to the user. Confirm all success criteria are covered before proceeding.
Launch 2-3 codebase-researcher agents in parallel via the Task tool:
Task(subagent_type="general-purpose", model="sonnet", run_in_background=true,
prompt="You are a codebase-researcher. First read *-context.md files, then dive into specific files. Research: {area}")
Each agent researches a different area relevant to the feature (e.g., data layer, UI layer, API layer).
Agents should FIRST read *-context.md files for progressive disclosure, then selectively read specific files.
Collect research findings: relevant files, patterns to follow, integration points, potential conflicts.
Launch the plan-writer agent with all gathered context:
Task(subagent_type="general-purpose", model="sonnet",
prompt="You are a plan-writer. Create a detailed implementation plan. Success criteria: {SC list}. Research: {findings}. Tests: {test structure}")
The plan must include:
Present the plan to the user for approval.
For each subtask from the plan, launch a general-purpose subagent:
Task(subagent_type="general-purpose",
prompt="Implement subtask N: {description}. Files: {paths}. Code: {snippets}. Success criteria: {relevant SCs}")
After each subtask completes:
just test and compare to the previous run (new passes? new failures? regressions?)just test-verbose output to fixAfter all subtasks complete:
just test -- all tests should pass (green phase)just test-verbose outputjust test-cov to check coveragestructural-completeness-reviewer agent for a final review