メニュー
Use when user wants to Creates a lightweight experiment document at .specs/sketches/{slug}.md to test a technical hypothesis before committing to a larger initiative.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Execute a Trellis task list by dispatching each task to a Claude Code subagent in its own context window. Use when the user says "build with sub-agents", "build with subagents", or when a feature has 5-20 tasks with dependencies between them. Reads `.specs/<feature>/tasks.md` produced by the tasks skill. Each subagent gets focused context (guidelines + plan + task + learnings) and returns results to the orchestrating session. Supports parallel execution of independent tasks via background subagents with worktree isolation.
Execute a Trellis task list using Claude Code's /batch command for bulk, patterned changes across many files. Use when the user says "build with batch" or when a feature involves repetitive changes across dozens or hundreds of files (e.g., migrating from one library to another, applying naming conventions, adding error handling patterns). Reads `.specs/<feature>/tasks.md` produced by the tasks skill. Translates the task list into one or more /batch invocations with worktree-isolated parallel agents.
Creates or updates .specs/guidelines.md with project-wide stack, conventions, and principles. Use when starting a project or when a fundamental technology decision has changed.
Use when user wants to Creates a pitch document at .specs/{feature}/pitch.md defining a problem domain with constraints, appetite, and solution shape. Use to frame a feature before writing a spec.
Use when user wants to Creates a technical plan at .specs/{feature}/plan.md translating a spec's functional requirements into architecture, technology, and code decisions.
Use when user wants to Creates a full functional specification at .specs/{feature}/spec.md defining what a system does and why. Use to write or revise feature requirements.
| name | trellis:sketch |
| description | Use when user wants to Creates a lightweight experiment document at .specs/sketches/{slug}.md to test a technical hypothesis before committing to a larger initiative. |
| allowed-tools | Task |
Create a new sketch document at .specs/sketches/<slug>.md.
Recommended effort: low. Short output; the user provides most of the content, the skill just structures it.
Run python3 ${CLAUDE_PLUGIN_ROOT}/scripts/validate-prereqs.py sketch and use the specsDir value from the JSON output. Abort if the output reports missing prerequisites.
If the user runs /sketch without enough context, ask:
If the user provides a description, extract the hypothesis from it. Every sketch must have a falsifiable hypothesis.
After gathering all user input, spawn the sketcher agent. Pass it: the specs directory path and all user-provided context (hypothesis, approach, constraints, slug name). The agent will generate the sketch document.
