---

Title: Managed AI Harness. Description: Deterministic Context v1 + managed plan/apply workflow for CAPYSQUASH.

MANAGED AI HARNESS

The engine and CLI provide deterministic artifacts. Managed AI orchestration (plan/apply) is handled by CAPYSQUASH managed services.

Command surface

Local OSS runtime does not expose ai-test, ai-demo, or ai-fix commands.

Use:

# deterministic context artifact

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

# managed feature discovery

capysquash features --json

Context v1 artifact

Default artifact name:

• .capysquash.context.v1.json

The artifact includes deterministic analysis context, candidates, ambiguous cases, and context_hash.

Managed endpoints

Submit Context/Plan through managed APIs in entitled environments:

• POST /api/harness
• POST /api/harness/handshake
• POST /api/harness/sessions/{session_id}/approve
• POST /api/harness/sessions/{session_id}/execute

Capability lookup:

• GET /api/features/ai-harness
• GET /api/harness/features/ai-harness (bearer-token flow)

Lock semantics

When managed harness is unavailable, responses return lock codes:

• FEATURE_LOCKED_LOGIN_REQUIRED
• FEATURE_LOCKED_SUBSCRIPTION_REQUIRED

Recommended flow

1. Run deterministic squash with - emit-context
2. Validate output with capysquash validate
3. Resolve managed feature capability
4. Start a managed session with the context and plan
5. Approve the reviewed session and execute with the returned single-use token

Notes

• Deterministic engine behavior remains available without managed services.
• Managed harness behavior is policy-gated by entitlement and plan.
• There is no client auto_approve flag.