← ClaudeAtlas

lh-tasklisted

Turn a rough idea or question into ONE clean, structured TODO.md item — grounded in the real codebase and refined with the user — ready to trigger the LHTask plan→implement→review hook chain. Use when capturing a task, triaging a "should we…?" question, or queuing work for the autonomous chain.
leonhoffmann86/lhtask-plugin · ★ 1 · AI & Automation · score 77
Install: claude install-skill leonhoffmann86/lhtask-plugin
You are turning a rough idea into **one well-formed task** in `TODO.md`. The rough idea: $ARGUMENTS This is a refinement workflow, not an implementation. **Do not write code.** Do not write to `TODO.md` until the task is crisp and the user has approved it. Work in these steps: ## 1. Read the constitution Find the project's constitution files and read them — obey their conventions, especially the **risk tiers** and any autonomous-agent / TODO-lifecycle rules. - If `lhtask.conf` exists, read the files listed in `LHTASK_CONSTITUTION_FILES` (in order). - Otherwise read whichever of `AGENTS.md`, `CLAUDE.md` exist (and any project-specific guide they reference, e.g. a frontend `CLAUDE.md`). - Skim `.githooks/README.md` if present, so you describe the downstream effect correctly. If none of these exist, tell the user the repo isn't bootstrapped yet and offer `/lhtask:bootstrap`. ## 2. Ground the idea in the real code (don't guess) Find the actual symbols/files the idea touches and its blast radius: - **Code graph** (preferred, if available): the `mcp__codegraph__*` tools (`codegraph_search`, `…_callers`, `…_callees`, `…_impact`, `…_context`), or the CLI (`codegraph context "<the idea>"`, `codegraph callers <symbol>`, `codegraph impact <symbol>`); run `codegraph sync .` first if the index looks stale. - **Fallback** (no code graph): use Grep/Glob to locate the relevant symbols and their callers. Read the few files that actually matter. Name concrete functions/files and wh