# 05

> 05 is the acceptance layer for AI-written code. It lets changes made by AI coding agents land with no pull request to review: the agent declares a plan, 05 seals it, checks the result on the exact code, signs the evidence, and keeps several agents on one codebase out of each other's way. A person approves by exception, and the change lands as a commit.

## Key facts

- Product name: 05 (domain use05.dev). Command line tool: `use05`. Open source under Apache-2.0: https://github.com/use05-dev/use05
- Install: `curl -fsSL https://use05.dev/install.sh | sh` (checksum-verified), or `npm install -g @use05/cli`, or build from source with Rust. macOS and Linux, x86_64 and arm64.
- Works with Claude Code and Codex through their hooks, with any other agent through a generic hook, and with every agent through a git pre-push gate and a CI gate.
- The loop: the agent declares its plan (`use05 start`), 05 seals it; checks run on the exact git tree and are signed as receipts (`use05 prove`); a deterministic gate compares plan and result; a person reads one page (`use05 brief`) only when something did not match and approves with a passphrase-protected key (`use05 approve`); `use05 land` puts the change on its target branch as a fast-forward commit; you push. Pull requests are optional.
- Several agents: each declares an area, others are told at the plan and at every edit, each works on its own branch and worktree, changes land one at a time, and finished branches and worktrees are removed.
- Folder and file rules in `.05/policy.toml`: `locked` paths are never changed (no approval clears them), `append_only` folders take new files only (for example database migrations), `no_delete` files may be edited but not removed, and `no_read` files are refused to agents' read tools (advisory). A person can bypass a rule for one exact change with a stated reason, unless the lock is `absolute`.
- Receipts are in-toto statements in DSSE envelopes, signed with Ed25519, verifiable offline with `use05 verify`.
- Gates are deterministic: no AI model judges a change. Local hooks are advisory, and 05 says so; a CI required check with a runner key the agent cannot read is the strong control.
- Privacy: the local tool runs entirely on your machine, with no account and no telemetry. 05 for teams, a paid product on the roadmap and launching soon, adds a simple web layer for managing how a team works in its code; it is designed to handle metadata only, never code.
- Not an agent, not an agent runner, not an AI code reviewer, not a Git host.

## Pages

- [05: the acceptance layer for AI-written code](https://use05.dev/): AI coding agents declare a plan, 05 checks the result and signs the proof, and approved changes land as commits. No pull request to review.
- [How 05 lands AI-written code without PR review](https://use05.dev/how-it-works): Agents declare a plan, 05 checks the exact code and signs the evidence, a person approves by exception, and the change lands as a commit. No PR needed.
- [Multiple AI agents on one codebase, without collisions | 05](https://use05.dev/multi-agent): Run several Claude Code or Codex agents on one repository: each declares its area, the others are told to stay out, and every change lands on its own.
- [Claude Code hooks and Codex hooks: agent guardrails | 05](https://use05.dev/agents): Claude Code and Codex hooks that seal the agent's plan, deny risky commands and protected paths, and keep parallel agents apart in their own worktrees.
- [Signed in-toto receipts for CI checks, verified offline | 05](https://use05.dev/receipts): 05 signs what a check did on an exact git tree as an in-toto statement in a DSSE envelope, the format SLSA uses. Anyone can verify it offline.
- [AI agent guardrails: what 05 enforces, and how strongly](https://use05.dev/enforcement): Agent hooks and the git pre-push hook are advisory; signed receipts and GitHub required status checks are strong. Where each 05 control runs.
- [AI agent governance: NIST AI RMF and ISO 42001 | 05](https://use05.dev/governance): How 05's sealed plans, signed receipts and approvals map to NIST AI RMF, ISO/IEC 42001 and the TBS guide on agentic AI for coding agents. Honest ratings.
- [Get started with 05: install, first task, CI gate](https://use05.dev/get-started): Install 05, run its first task and read the brief, then turn on the CI gate that makes it enforceable. About five minutes, with real terminal output.
- [05 FAQ: no PR review, Claude Code hooks, privacy](https://use05.dev/faq): Do you still need pull requests? Is 05 an AI code review tool? How do its Claude Code and Codex hooks work? Straight answers about 05.
- [About 05: the trust layer for agent-written software](https://use05.dev/about): Why we are building 05, what it does today and what it will not do: an open-source acceptance layer for AI-written code.
- [Privacy: anonymous analytics, no cookies, no telemetry | 05](https://use05.dev/privacy): This site counts page views anonymously, sets no cookies and loads nothing from third parties. The 05 local tool has no telemetry. Here is what we see.

## Source and docs

- [Source code and README](https://github.com/use05-dev/use05): the CLI, runner and receipt format, Apache-2.0
- [Turn on the CI gate](https://github.com/use05-dev/use05/blob/main/docs/turn-on-the-ci-gate.md): the strong control
- [Receipt format](https://github.com/use05-dev/use05/blob/main/docs/receipts.md): every record type and how to verify it
- [Extending 05](https://github.com/use05-dev/use05/blob/main/docs/extending.md): commands, any agent's hooks, events, your own record types
- [Releases](https://github.com/use05-dev/use05/releases): checksummed binaries for macOS and Linux

## Optional

- [Full text of every page](https://use05.dev/llms-full.txt)