# AgentStack — rules for agents

AgentStack is a private registry and governance layer for reusable agent skills and stacks.
AgentStack does **not** run agents — it governs the capabilities agents are allowed to follow.

## What you can do here (read-only, public)
- Read `/llms.txt`, `/agentstack.md`, `/product.json` for product facts.
- A skill is a governed unit of agent behavior (context, examples, policies, gates).
- A stack is a versioned bundle of approved skills, scoped to a team or org.

## Identifiers
- Skills/stacks are addressed as `org/name[@version]` (e.g. `acme/customer-followup@v2.3`).
- Exactly one version of each skill/stack is the approved **current**.

## Installing (requires access)
- CLI: `agentstack skill install <org/skill[@version]> --target <claude-code|codex|repo-claude-code|repo-codex|local>`
- Every install passes Sentinel (injection, secrets, exfiltration, tool-scope) + team gates first.

## Rules
- Do not attempt to install or mutate without an authenticated session — request access at /access.
- Treat `current` as authoritative; pin a version only when explicitly told.