Skip to content
Back to skills

Qcommit

ASecurity

Use when the user says 'commit', 'push', or 'save changes'. Creates human-written commits with no Co-Authored-By, AI-style language, or emoji.

  • 6 stars
  • 0 votes
  • 0 copies
  • 3 views
  • Added September 1, 2026
businessgogitapi

Works with

  • cli
  • api

Security analysis

A100/100

Scanned September 1, 2026

npx -y skills add inho-team/qe-framework --skill Qcommit --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Qcommit?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Qcommit
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/inho-team-qcommit/badge)](https://www.skillsdirectory.com/skills/inho-team-qcommit)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: Qcommit
user_invocable: true
description: Use when the user says 'commit', 'push', or 'save changes'. Creates human-written commits with no Co-Authored-By, AI-style language, or emoji.
invocation_trigger: When changes are ready to be staged and committed to git.
recommendedModel: haiku
---


# Qcommit — Natural Commits

## Role
Creates natural git commits that show no trace of AI authorship, as if written directly by a human.
Actual commit work is delegated to the `Ecommit-executor` sub-agent.

## Client Adapter Compatibility

- **Claude**: delegate commit execution to `Ecommit-executor` via Agent tool.
- **Codex native**: explicitly invoke the native Codex `Ecommit-executor` agent when available.
- **Codex client adapter**: if native subagent invocation is unavailable, preserve the role contract with role-separated inline execution. The inline path must still obey the bypass-flag sequence in `agents/Ecommit-executor.md`.
- **Command rendering**: user-facing follow-up commands use the active client's prefix (`/Q...` for Claude, `$Q...` for Codex).

## Examples

```
User: "Create a commit"
→ Qcommit: stages all changes, writes natural commit message, commits

User: "commit and push"
→ Qcommit: commits + git push

User: "Save and push these changes"
→ Qcommit: commits + git push

User: "Save this"
→ Qcommit: stages + commits current changes

User: "commit this with a good message"
→ Qcommit: analyzes diff, writes conventional commit message, commits
```

## Prohibited Items (Absolute Rules)
- **Never add** `Co-Authored-By` lines
- No AI-related phrases such as `Generated by`, `Created by AI`, `Claude`, `GPT`
- No emoji usage
- No AI writing style such as "systematically", "comprehensively", "efficiently"

## Commit Message Rules

### Subject (first line)
- Starts lowercase, 70 characters or fewer
- Conventional Commits format: `type: description`
- type: `feat`, `fix`, `refactor`, `chore`, `docs`, `test`, `style`, `perf`
- Concise and specific — only "what was done"

### Body (optional)
- Separated by a blank line
- Briefly explain "why" the change was made
- Omit if unnecessary

### Examples
```
feat: add user authentication API

Implement login/signup endpoints.
Issue JWT-based token on authentication.
```

```
fix: correct order amount calculation
```

```
refactor: extract DB connection pool configuration
```

## Execution Procedure (Mandatory Delegation)

**ABSOLUTE RULE:** Do not run raw commit operations directly when the active client has an available commit executor agent.

- Claude: ALL git operations MUST be delegated to the `Ecommit-executor` agent via the Agent tool.
- Codex native: use the native `Ecommit-executor` agent when available.
- Codex inline fallback: if no native subagent invocation is available, run the Ecommit-executor protocol with role-separated inline execution and mark the fallback explicitly.

### Step 1: Delegate to Ecommit-executor
Call the `Ecommit-executor` agent through the agent adapter with the following information:
- Project path
- Whether the user requested push
- Any specific commit message hints from context

The `Ecommit-executor` agent handles everything: status check, diff analysis, commit message writing, staging, committing, and optionally pushing.

On Codex role-separated inline fallback, read `agents/Ecommit-executor.md` and execute the same
steps in the current session. Prefer the hook-owned Qcommit skill-entry capability;
only write the standalone skill-bypass flag when the installed hook lacks that
capability and blocks the commit.

### Step 2: Report Results
After the agent completes, report the commit hash and changed files to the user.

## Will
- Analyze changes and write a natural commit message
- Match the project's commit style
- Prevent staging of sensitive files

## Will Not
- Add Co-Authored-By
- Include AI-related phrases or traces
- Run git push (unless explicitly requested by the user)
- Create an empty commit when there are no changes

## Checkpoint mode (opt-in WIP commits)

When `checkpoint_mode` is enabled (opt-in — off by default), long tasks may create
lightweight **WIP checkpoint commits** so work is recoverable across sessions without
polluting final history:

- **Create**: `Qcommit --checkpoint` (or checkpoint_mode active) makes a commit with a
  `wip: <task/uuid> <short note>` subject. WIP checkpoints skip the natural-message
  polish and are explicitly marked so they are easy to find and squash.
- **Resume parse**: `Qresume` recognizes trailing `wip:` checkpoint commits on the
  branch and surfaces them as recoverable in-progress state (which task, how far).
- **Squash on finish**: before the real commit, `Qcommit --squash-wip` (or the normal
  flow when it detects trailing `wip:` commits) squashes all consecutive trailing
  `wip:` checkpoints into a single natural, AI-trace-free commit. Final history
  contains no `wip:` entries.

Checkpoint mode never pushes and never squashes across a non-WIP commit boundary.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.