action

Generate categorized changelog entries from branch commits and PR context; refresh or recut a branch's own entries as it evolves

Run prompts through individual AI CLIs — Antigravity/agy (Gemini), codex/GPT, and cursor/agent with fallback support

Scan repo tracked files for AI slop, brittle counts, and verbose noise; resolve each finding as an atomic forward-going commit with quality-gate verification

Write tight, slop-free prose and code, and tighten existing files in place with no commits — a model-invocable writing discipline plus a working-tree cleanup command

TDD bug-fix workflow — reproduce bugs as failing tests, find root cause, fix, and verify

Address code-review findings on the current branch — provenance-gated to changes the branch introduced, one finding per commit, simplest pragmatic fix, with quality gates before every commit and prompted history rewrites

Run independent adversarial participants through host-native sub-agents by default or separate model CLIs by choice — brainstorm, refine, plan, execute, architect, review, and synthesize

Generate, refresh, recut, and review gold-standard PR descriptions; detect AI slop, brittle counts, and verbose commit messages on branch commits and resolve via fixup commits and autosquash with quality-gate checks

Merge PRs via gh with merge commits matching the repo's own git history — readiness checks and CI watch before every merge, plus multi-PR runs with stack detection, rebase, and conflict resolution between merges

Gain situational awareness before touching a repository — read the branch against trunk, its diff, its pull request and review threads, its linked tickets, and the project's own conventions, then report where the work stands; optionally search prior AI conversations for decisions the repository never recorded