tobihagemann/turbo / claude/skills/write-skill
write-skill
VerifiedCreate or update a skill that extends Claude's capabilities. Use when the user asks to "create a skill", "make a new skill", "write a skill for...", or "update this skill".
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 9.9 KB |
| references/composition.md | markdown | 4.6 KB |
| references/evaluation.md | markdown | 3.9 KB |
| references/harness.md | markdown | 2.9 KB |
| references/principles.md | markdown | 7.3 KB |
| references/scripts.md | markdown | 4.4 KB |
| references/skill-reviewer.md | markdown | 5.7 KB |
| references/structure.md | markdown | 9.4 KB |
| references/tools.md | markdown | 28.1 KB |
| references/workflows.md | markdown | 8.0 KB |
| references/writing.md | markdown | 8.1 KB |
references/harness.md
View on GitHub# Harness Reference
Quick reference for the Claude Code harness: canonical tool names, limits, and discovery paths.
**Tools (canonical names):**
- `TaskCreate` / `TaskUpdate` / `TaskList` / `TaskGet` — task tracking. Statuses: `pending`, `in_progress`, `completed`, plus `deleted` for removal. Tasks support `owner`, `blocks` / `blockedBy`, and arbitrary `metadata`.
- `AskUserQuestion` — structured user prompts. 1-4 questions per call, 2-4 options per question, a header of at most 12 characters, option labels of 1-5 words, "Other" free-form option appended automatically.
- `Edit` / `Write` / `Read` / `NotebookEdit` — file operations. `Edit` requires the file to be read first.
- `Bash` — shell execution; supports `run_in_background` for long-running commands.
- `Skill tool` — invoke an installed skill by name.
- `Agent tool` — spawn a sub-agent. The `subagent_type` field selects a specialized agent (built-ins include `Explore`, `Plan`, `general-purpose`, plus user-defined types). Generic sub-agents accept a `model` override. The `name` field makes the sub-agent an addressable teammate, which changes how its result is delivered; see Never Name a Spawned Agent in `tools.md`.
- `SendMessage` — message another agent by name, or by the raw `agent_id` from its spawn result when it has no name. Background sub-agents can also message `main`.
**Sub-agents:** No documented hard cap on count or nesting; pragmatic limits are whatever the parent context can manage. Sub-agents inherit the parent model unless `model` is specified. Specialized agents have restricted tool subsets defined per agent type.
**Permission modes:** `default`, `acceptEdits`, `plan`, `bypassPermissions`, `dontAsk`. Set via `--permission-mode`, `/permissions`, or per-tool allow-lists. `plan` mode blocks writes until plan approval.
**Skill discovery:**
- User-scope: `~/.claude/skills/`.
- Project-scope: `.claude/skills/` (cwd).
- Plugin-scope: a plugin's `skills/` folder, named `<plugin>:<skill>`. Typed by the user, a plugin skill resolves only as `/<plugin>:<skill>`. Invoked from a sibling skill in the same plugin, the bare `/<skill>` name also reaches it, and a same-named user or project skill takes that bare name. `${CLAUDE_SKILL_DIR}` in a plugin `SKILL.md` expands to that skill's own folder.
- Skills invoked as `/skill-name`; `SKILL.md` frontmatter requires `name` and `description`.
**CLAUDE.md hierarchy:** Auto-loaded from each ancestor of the cwd up through the home directory. Project-root `CLAUDE.md` is the canonical project instruction file; `~/.claude/CLAUDE.md` is the user-global file.
**Hook system:** `PreToolUse`, `PostToolUse`, `UserPromptSubmit`, `Stop`, and others. Configured under `hooks` in `~/.claude/settings.json` or `.claude/settings.json`.
**Config:** `~/.claude/settings.json` (user) and `.claude/settings.json` (project, with `.local` variants for gitignored overrides).
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |