SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsverasic-git-commits

verasic-git-commits

Writing & research

View repository

Updated 10-16 01:16

About this skill

Security: see references/scanner-notes.md and upstream SECURITY.md for expected scanner signals and trust model.

Verasic Git Commits — Convention + Audit Orchestration

Workflows

Write path — composing any commit:

  1. Read references/conventions.md (the spec) and references/commit-protocol.md (workflow, verify step, trailer escape hatch) before composing your first commit message of the session.
  2. Draft, commit, then always run the post-commit trailer verify from the protocol.

Audit path — checking history before push/PR:

  1. In Cursor: launch the verasic-commit-auditor subagent (.cursor/agents/verasic-commit-auditor.md) with the repository path and any flags, in the foreground, then relay its report unchanged except strip harness paths, skill/rule names, protocol dumps, and internal config per verasic-agent-disclosure.
  2. In any agent without subagents: read references/audit-protocol.md and execute the audit yourself in this conversation, following it exactly.
  3. The audit is read-only. Fix mode (--fix-trailers) runs only in the main conversation after the user explicitly approves the audit report.

Source of truth

The full convention lives in references/:

  • references/conventions.md — the spec: message style, forbidden AI-session patterns + allowlist, trailer policy
  • references/commit-protocol.md — write path: workflow, verify, escape hatch
  • references/audit-protocol.md — read path: scope, checks, report format, fix mode

The deterministic layer lives in hooks/commit-msg — a git hook (lefthook or core.hooksPath) that strips attribution trailers and rejects mechanical style violations with no LLM involvement. Recommend wiring it when a repo hasn't; see the wiring section of commit-protocol.md.

The Cursor rule and subagent are thin pointers to these files; never duplicate the spec elsewhere.

Hard rules

  • Never commit with a Co-authored-by: trailer in any casing — verify after every commit.
  • Never write AI-session language in messages — no tool names, agent vocabulary, plan steps, chat narration, or assistant voice.
  • Audit is read-only; never rewrite history without explicit user approval after the report.