mit einem Klick
init-claude-md
Triggers on /init-claude-md only.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Menü
Triggers on /init-claude-md only.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Basierend auf der SOC-Berufsklassifikation
Triggers on /clockify-reconciliator <project> only. Adds descriptions to description-less Clockify entries for a configured project, splitting large blocks into 1-3h chunks using git commits from configured repos.
Triggers on /work-recap only. Dispatches to a named recap variant (e.g. zirtue weekly, zirtue daily). Each variant lives in a subfolder under this skill.
Triggers on /schedule-once only. Schedules a SINGLE future run of a Claude prompt (or a raw shell command) on THIS machine via Windows Task Scheduler, as a self-deleting one-time task. The local, one-shot counterpart to the cloud /schedule and the recurring /cron-run. PC must be on and logged in at fire time; Claude Code itself need not be open. Also supports list and cancel.
Triggers on /create-pr only. Drafts a human-light PR for the current branch, scales the body to the diff, suggests visuals, previews locally, and creates it on approval.
Triggers on /autopilot only. Dev is AFK and wants maximum autonomous progress with heavy but purposeful token use. Never block - delegate aggressively to subagents to keep main context lean, resolve real judgment calls via a BOUNDED /iterate-it (capped per run), auto-answer any nested skill's question instead of hanging, log decisions, park true hard-stops, and grind the task to a verified finish.
Triggers on /batch-todos only. Dedupes ai_todos, classifies survivors as EASY (auto-execute) or HARD (dev picks), shows dry-run confirmation, batches all EASY todos, then surfaces the HARD queue.
| name | init-claude-md |
| description | Triggers on /init-claude-md only. |
Generate or update the project-level CLAUDE.md file.
Reader: Claude Code, not a human. Optimize for Claude's effectiveness, not human brevity. Include anything that takes multiple files to discover. Omit only what Claude can trivially find in one read (e.g. a single config file).
If the user passed skipVerification, skip this step entirely and proceed to Step 1.
If CLAUDE.md exists and already has the correct structure (Project section with description + Type + Deploy, Structure section, Commands section, Rules section - Architecture section optional), print:
/init-claude-md - already complete, skipping.
And stop. Only proceed if CLAUDE.md is missing or malformed.
If it exists, read it. The goal is to update it to match the standard structure below while preserving anything that looks intentional and project-specific.
If it does not exist, create it from scratch.
Read enough to confidently fill the standard fields and understand the architecture:
package.json, vite.config.*, index.html, pubspec.yaml, default.project.json or equivalent to determine project type.github/workflows/ to determine deploy setup.prettierrc, .gitignore to see what boilerplate existspackage.json scripts section to find dev/build/test commandsDo not read every file. Answer: what type is this, how is it structured, how does it deploy, how do you verify changes, and what are the subsystems a new Claude session would need a map to navigate.
Write or update the file at the project root. Follow this structure exactly:
## Project
One sentence: what this app does and who it's for.
Type: html | vite | react | roblox | flutter | other
Deploy: github-pages | none | other
## Structure
Brief one-liner of the source layout, e.g. "src/styles.css, src/script.js, assets/images/, assets/fonts/, assets/data/"
## Commands
- Dev: `<dev command>`
- Verify: `<build or type-check command>`
- [only include if project has a test command] Test: `<test command>`
## Architecture
[For simple projects: omit this section entirely.]
[For projects with multiple subsystems, non-obvious module boundaries, or cross-cutting flows: describe them here. Include file maps, data flow, and anything that takes more than one file to understand. This is the highest-value section for Claude - a session without this map re-discovers the architecture every time.]
## Rules
- [any project-specific rules that are genuinely useful]
| Value | When to use |
|---|---|
html | Plain HTML/CSS/JS, no build step |
vite | Vite project, no framework |
react | React project (Vite or CRA) |
roblox | Roblox Luau project |
flutter | Flutter/Dart project |
other | Anything else |
Only add rules that are genuinely project-specific and that Claude would not know without being told. Good examples:
Do not add rules that are already in the global CLAUDE.md or that are obvious from the project type.
No line limit. Every line must earn its place by this test: "would Claude need to open a file to know this?" If yes, include it. If no, omit it. Do not repeat what is already in the global CLAUDE.md or obvious from file names and project type.
Tell the user what was written or updated and flag any assumptions made. Do not commit - the user handles that.