agentic-os
Governance framework for AI coding agents. It runs them through a five-step workflow (plan, build, review, test, ship) where no step counts as done without evid
What is agentic-os?
Governance framework for AI coding agents. It runs them through a five-step workflow (plan, build, review, test, ship) where no step counts as done without evidence. Drop-in rules and guardrails for Claude Code, Codex, Cursor, Copilot, and Antigravity, via AGENTS.md.
Overview of agentic-os
What it is
Agentic OS is a governance framework for AI coding agents. It enforces a five-step workflow (plan, build, review, test, ship) where each step requires evidence before proceeding. The framework provides drop-in rules and guardrails for agents like Claude Code, Codex, Cursor, Copilot, and Antigravity via AGENTS.md files. It includes a credential scanner, a work trail validator, and git hooks that run independently of the agent's report.
Key features
- Machine-enforced backstops: Credential scanning and test validation run in git hooks and CI, not relying on the agent's word. A local validator checks the work trail for missing phases or evidence.
- Memory that survives sessions: Each task maintains a work log of decisions, evidence, and gate receipts. The
/handoffcommand writes a resume block so the next session or agent can continue from the log. - 14 expert checklists attached by task type: Checklists like auth-security for login code or forward-only checks for migrations are automatically associated with the relevant phase.
- Gated phases scaled to risk: Workflow paths vary by classification (tiny-fix, quick-win, feature, hotfix, architecture-change), each with required phases. Skipping a phase causes
validate.shto fail. - Cross-platform: One set of governance files works across major AI coding agents.
Use cases
- Preventing leaked secrets from reaching git history by running a credential scan before commits.
- Ensuring AI agents follow a defined workflow (plan, build, review, test, ship) with evidence for each step.
- Maintaining context across agent sessions or agent handoffs via persistent work logs.
- Enforcing test execution and code fingerprinting to detect stale or failed test runs.
Getting started
To run a credential gate demo without installation:shbash demo/run.sh # Windows (PowerShell): pwsh demo/run.ps1For full installation, see the repository README. The installer prints the commands to enable the pre-commit hook and CI job.
License and pricing
License: MIT. Pricing: open source.
Based on the project's public README and repository metadata.