Menu
Diff staged/unstaged changes, write a conventional commit message, and commit immediately — no confirmation needed.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Avoid filler words like "actually", "wait", "hmm", and other verbal stutters when thinking, narrating, or reasoning. Flow directly from thought to thought instead of announcing self-corrections aloud. Use whenever writing internal thinking, problem-solving narration, or any prose where you catch yourself about to say "Wait" or "Actually" mid-thought.
Given one or more GitHub repo URLs, perform a thorough analysis of code quality, AI usage, open issues/PRs, and install feasibility, producing a structured report.
Operate the user's tldraw offline canvas app, including open .tldraw or .tldr files. Use whenever a task involves inspecting, editing, arranging, connecting, linting, or scripting a tldraw Desktop canvas.
(no description)
Based on SOC occupation classification
| name | draft-commit |
| description | Diff staged/unstaged changes, write a conventional commit message, and commit immediately — no confirmation needed. |
| triggers | ["draft a commit","draft commit"] |
You are executing the draft-commit skill. If you are running in Claude Code and can spawn a subagent via the Agent tool, do so using Haiku to keep cost low. If the Agent tool is unavailable or model spawning is not supported in your environment, skip that step and execute the prompt below directly yourself.
Before building the agent prompt, check whether the user's request includes phrases like "staged changes", "what I've staged", "already staged", or "staged only". If so, use staged-only mode. Otherwise use default mode.
If spawning a subagent is supported, call the Agent tool with these exact parameters:
subagent_type: omit (use general-purpose)model: "haiku"description: "Draft and commit changes"prompt: the full prompt below for the detected mode (substitute the actual working directory path if you have it, otherwise let the agent discover it)You are committing all current changes in a git repository. Do this immediately — no confirmation needed.
Working directory: (the current working directory of the parent session)
Steps:
rtk git diff HEAD — read the full diffrtk git status — see untracked files (never use -uall)rtk git log -5 --oneline — match the repo's commit stylertk git add -u
noreply@anthropic.com for Anthropic models, noreply@deepseek.com for DeepSeek models, noreply@google.com for Google models, and so on). If you are uncertain, use noreply@ai.local as a fallback.feat, fix, chore, refactor, style, docs, perf, or testCo-Authored-By: <your model name> <appropriate noreply email>rtk git commit -m "$(cat <<'EOF'
<title>
<body if needed>
Co-Authored-By: <your model name> <appropriate noreply email>
EOF
)"
Do not push. Do not ask for confirmation.
You are committing only the already-staged changes in a git repository. Do NOT stage anything new. Do this immediately — no confirmation needed.
Working directory: (the current working directory of the parent session)
Steps:
rtk git diff --cached — read only the staged diffrtk git status — confirm what is staged vs unstaged (never use -uall)rtk git log -5 --oneline — match the repo's commit stylegit add — commit exactly what is already staged.noreply@anthropic.com for Anthropic models, noreply@deepseek.com for DeepSeek models, noreply@google.com for Google models, and so on). If you are uncertain, use noreply@ai.local as a fallback.feat, fix, chore, refactor, style, docs, perf, or testCo-Authored-By: <your model name> <appropriate noreply email>rtk git commit -m "$(cat <<'EOF'
<title>
<body if needed>
Co-Authored-By: <your model name> <appropriate noreply email>
EOF
)"
Do not push. Do not ask for confirmation.