agent-init

To create an AGENTS.md file, audit your repository for non-discoverable guidance like command quirks, landmines, and conventions. Scan configs, docs, and CI to build an evidence-backed profile, then draft incremental diffs that pass a quality gate before writing.

AGENTS.md files should contain only hidden, high-signal guidance that agents cannot infer from README, code, or tooling. This includes missing command flags, undocumented landmines, repository conventions, and routing needs that add measurable value to AI agent onboarding.

Detect missing constraints and stale instructions by running a guided gap detection phase that scans configs, docs, CI, and tooling. This intelligence highlights missing command flags, landmines, conventions, and routing needs while preventing redundant or discoverable content from being included.

The best way to onboard an AI agent is to audit the repository for non-discoverable constraints and draft a minimal AGENTS.md file. Apply a discoverability filter to ensure every instruction adds measurable value and log evidence before writing the final file.

Yes, you can update an existing AGENTS.md file using incremental diffs. The safety-first drafting phase enforces a discoverability filter and quality gate, ensuring updates are minimal, accurate, and actionable without overwriting existing valid instructions.

AI agents trigger build failures when repository conventions and command landmines are non-discoverable. Auditing the repo to surface these hidden constraints into an AGENTS.md file captures the high-signal guidance needed to prevent errors and ensure safe execution.