---

Title: AI Integration. Description: Managed AI harness integration for deterministic migration workflows.

AI INTEGRATION (MANAGED)

CAPYSQUASH uses a deterministic local engine plus a managed harness for entitlement-gated AI orchestration.

Architecture

• Local/OSS: parse, analyze, squash, validate, emit deterministic Context v1.
• Managed: capability checks, plan validation, approval gating, governed execution.

Local commands

capysquash analyze migrations/
capysquash squash migrations/ --emit-context --context-output ./squashed/.capysquash.context.v1.json
capysquash validate migrations/ squashed/
capysquash features --json

Managed endpoints

Use managed APIs from entitled clients:

• GET /api/features/ai-harness
• POST /api/harness
• POST /api/harness/handshake
• POST /api/harness/sessions/{session_id}/approve
• POST /api/harness/sessions/{session_id}/execute

Bearer-token managed session flow:

• GET /api/harness/features/ai-harness
• POST /api/harness/workflow
• POST /api/harness/handshake
• POST /api/harness/sessions/{session_id}/approve
• POST /api/harness/sessions/{session_id}/execute

Contract notes

• Canonical context artifact: .capysquash.context.v1.json
• Canonical hash field: context_hash
• Lock codes: FEATURE_LOCKED_LOGIN_REQUIRED, FEATURE_LOCKED_SUBSCRIPTION_REQUIRED
• There is no client auto_approve flag. Review-required sessions must call approve and then execute with the returned single-use approval token.

Migration from older docs

Legacy references to local runtime commands (ai-test, ai-demo, ai-fix) are deprecated from OSS command surfaces.