Action-safety bridge for Company Brain context

[100yenadmin]

Define the plugin/API bridge for action safety: can this agent act on this context, does this require approval, is the evidence stale/contradicted/sensitive, and what should be shown to the operator.\n\nAcceptance criteria:\n- Runtime can request action-readiness metadata from Cortex.\n- Approval-gated context is not silently acted on.\n- Operator-facing explanation includes source/provenance when available.\n\nRoadmap: /Users/lume/.openclaw/workspace-eva/docs/projects/company-brain/plan-company-brain-ai-os.md

[100yenadmin]

Roadmap sync for May 8, 2026: Cortex Company Brain now surfaces pilot action-readiness metadata (visibility_scope, requires_approval, action_readiness, verification_status) through account brief/timeline/query responses.

This issue remains the action-safety roadmap anchor. Concrete tool plumbing is now tracked in #26, with upstream hardening in electricsheephq/electric-sheep#2032 and fixtures in electricsheephq/electric-sheep#2033.

Implementation boundary: approval-gated context can be shown, drafted from, or escalated, but must not silently trigger autonomous outbound action.

Upstream epic: electricsheephq/electric-sheep#2029.

[100yenadmin]

Planning alignment update.

This should pair with #24 and dashboard#156. The plugin/runtime must preserve Company Brain action-readiness metadata and prevent approval-gated context from becoming silent outbound action. The dashboard queue issue is electricsheephq/electric-sheep-website-dashboard#156, and the staged canary is electricsheephq/electric-sheep#2040.

[100yenadmin]

Activation slice landed in #29 and dashboard queue support landed in electricsheephq/electric-sheep-website-dashboard#161.

The plugin preserves requires_approval, action_readiness, verification_status, visibility_scope, citations, and insufficient-evidence metadata, and exposes an action-safety envelope with no executable outbound actions.

Keeping this issue open as the broader action-safety roadmap anchor for production policies and future operator/runtime enforcement work.

[100yenadmin]

Sprint update: action-safety metadata now flows through the explicit Company Brain tools/context path: citations, insufficient_evidence, requires_approval, action_readiness, verification_status, and visibility_scope are preserved for operator review, and PR #30 fixed packaged manifest parity for the context settings. No outbound action execution was added. Closing as satisfied for the pilot bridge; broader autonomous action policy remains out of scope.