Analysis updated 2026-05-18
Consolidate a project's CLAUDE.md and AGENTS.md into a single canonical file so AI agents always read accurate guidance.
Build a searchable log of past bugs and gotchas in docs/solutions/ so AI agents do not repeat the same mistakes across sessions.
Add a pre-commit secret scan and session startup hooks to a repository where multiple AI agents are making commits.
| anmoln7/agent-standard-oss | trivo25/code-airlock | chrisor-dev/claude-autosync | |
|---|---|---|---|
| Stars | 5 | 7 | 2 |
| Language | Shell | Shell | Shell |
| Setup difficulty | easy | hard | moderate |
| Complexity | 2/5 | 3/5 | 3/5 |
| Audience | developer | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Docs-only adoption requires only cloning the repo and renaming files, automated features optionally add shell scripts and git hooks.
Agent Standard is a convention for keeping the instruction files that AI coding agents read, usually named AGENTS.md or CLAUDE.md, accurate and up to date as a codebase changes. These files tell agents how to work in a repository: what commands to run, what rules to follow, what pitfalls to avoid. The problem the project addresses is that these files drift. When a codebase has two instruction files, or when a README also documents some of the same rules, they fall out of sync and the agent starts following stale guidance confidently. The core of the standard is a small set of conventions. One file, AGENTS.md, is the single authoritative source. If you also have CLAUDE.md (the file Claude Code reads), it becomes a one-line reference pointing at AGENTS.md. A docs/solutions/ directory holds a separate file for each past bug or gotcha the team has encountered, making it searchable. A designated section in the main instruction file lists file pairs that must always agree, creating an explicit anti-drift contract that can be checked automatically. Beyond the documentation conventions, the project includes optional shell scripts and hooks for safer day-to-day work with AI agents: a pre-commit scan for secrets, session startup hooks that silently repair common configuration failures before they cause problems, commit identity rules, helpers for safe landing of changes, and a multi-account deploy guardrail. Adoption comes in layers. The docs-only approach requires no scripts and no automation: clone the repo, rename your instruction file, create the fix log directory, and run a read-only check. The fuller adoption adds shell scripts to your PATH and optional git hooks. An interactive wizard called adopt guides you through either path with a before-and-after scorecard and a question before every change. Nothing is deleted automatically. The standard works with Claude Code, Codex, Cursor, and Gemini. The AGENTS.md file is the format those tools read natively. A Claude Code plugin wraps the wizard in a slash command. The project is MIT licensed and includes a CI action that checks compliance automatically.
A convention and toolset for keeping AI agent instruction files (AGENTS.md, CLAUDE.md) accurate and single-sourced across a codebase, with anti-drift contracts, a fix log, and safe commit helpers.
Mainly Shell. The stack also includes Shell, Bash, Git.
MIT, free to use, modify, and distribute for any purpose including commercial.
Setup difficulty is rated easy, with roughly 5min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.