[KAVentures]

Purpose

Ship the first small, backward-compatible slice of map workflows → feed agents → make AI do the work, without turning OWG into an RPA platform.

Changes

• Add pure, bounded shared/agent_workflow_map.py projection over the existing authorized evidence bundle.
• Include additive agent_workflow_map in /v1/workflow-evidence, MCP get_workflow_evidence, and existing exports. No new route, grant, background sync, database schema, or agent execution surface.
• Expose support-counted observed sequence steps, variations, opaque execution refs, known truncation/coverage gaps and missing business rule questions.
• Update MCP guidance for external agents to consume the map, use their existing authorized tools, and independently verify real source-state outcomes.
• Add tests for deterministic ordering, privacy boundaries, truncated/inferred observations, no implicit policy/action authority, and existing bundle integration.
• Document workflow handoff and the real A/B/C outcome validation still needed.

Guardrails

Default privacy and history-access behavior unchanged. No raw new capture, no automatic execution, no assumed policy, no clipboard contents, no reliance on timing matches as proof. Canonical evidence overrides a derived map.

Validation

CI requested by opening this PR; no local checkout/test run available in this execution environment. Please review CI before merging.

Next

Real-world agent trials comparing application APIs alone vs +logs vs +OWG maps/evidence; score independently verified task completions, unnecessary interventions, false claims, latency, and setup costs. No unsupported product-uplift claims.