Glossary
Terms for contract drift monitoring, the status plane, and agent control surfaces.
- Watch
- A monitored external contract — one REST/OpenAPI URL or MCP endpoint on a schedule. Watches →
- Snapshot
- Point-in-time capture of a watch target (JSON schema, OpenAPI doc, or MCP
tools/list). Baseline snapshots anchor future diffs. - Fingerprint
- Stable hash of normalized schema shape. Identical fingerprints skip redundant diff work when content unchanged.
- Change record (drift event)
- Created when a new snapshot differs from the previous baseline. Contains classified
changes[]with severity. Change records → - drift_status
- Canonical watch health:
ok,drifted,error,never_run, ordisabled. Used by console chips, preflight, and MCPget_watch_status. - Breaking change
- Contract change that can break strict clients or agents — removed tools, removed required fields, operation removal. MCP tool removal is always breaking.
- agentAction
- Structured remediation hint on a drift change (e.g.
update_tool_schema,ack_incident) — for humans and orchestrators. - Incident
- Open drift state on a watch until acknowledged or remediated. Policies may require ack before new agent runs resume.
- Agent binding
- Maps a production agent (id/slug from
agents.yaml) to watches and a drift policy. CP-2 control-plane object. - Pre-run check (preflight)
- Ask before you run:
POST /api/preflightor MCP status tools returnallowedor blocked watches with reasons. Pre-run check guide → - Coverage assert
- CI check that discovered dependency URLs in a repo are watched. Coverage →
- FuseGuard
- OSS loop fuse that trips on repeated tool failures; can integrate hosted preflight before outbound calls. How-to →