A pack is the unit of agent-ready knowledge.
Every pack bundles six artifact classes — each with its own loading mode — plus activation metadata that tells an agent when the pack should fire.
Reference material fetched only when an agent asks.
Always-on constraints injected into the agent's working context.
Pushed when an activation trigger fires for the current task.
Loaded by the validator/doorkeeper when an action is proposed.
Loaded by the eval runner to score baseline vs. AFPS-augmented behavior.
Emitted as cryptographic evidence of what the agent actually did.
Pack lifecycle
- 01Build
Author the pack: rules, workflows, schemas, tests, and activation metadata.
- 02Evaluate
Run scenario evals comparing baseline vs. with-pack behavior. Fail the gate below threshold.
- 03Distribute
Publish to a registry. Each version gets a scorecard, signature, and provenance.
- 04Operate
Agents install, activate on triggers, emit receipts, and feed telemetry back.
Featured packs
Browse the registry →- AFPS-4
Agentic Privileged SaaS Access Flow
Classify, approve, gate, execute, receipt, and revoke privileged SaaS actions performed by AI agents.
version0.4.1safety96scenario coverage92afps install agentic-privileged-saas-access-flow - AFPS-3
Secure PR Review
Force agents to attach SAST + dependency-risk context before approving a pull request.
version0.2.0safety90scenario coverage81afps install secure-pr-review - AFPS-3
Incident Runbook v2
Push the on-call runbook + paging rules to any agent acting during a P0/P1 incident.
version0.3.4safety88scenario coverage74afps install incident-runbook-v2 - AFPS-2
PII Handling (EU)
Eager rules + validation schemas for GDPR-bound data flows.
version0.1.7safety93scenario coverage67afps install pii-handling-eu