Company Brain injection mode

[100yenadmin]

Add a plugin/runtime mode that distinguishes personal memory, organization memory, customer/account context, and source-cited operating state when injecting Cortex context into agent turns.\n\nAcceptance criteria:\n- Injection labels context scope clearly.\n- Source-cited operating state can be included without flattening into personal memory.\n- Non-authoritative GBrain/shadow context remains marked as such.\n\nRoadmap: /Users/lume/.openclaw/workspace-eva/docs/projects/company-brain/plan-company-brain-ai-os.md

[100yenadmin]

Roadmap sync for May 8, 2026: the Cortex backend now exposes the Company Brain account/list/brief/timeline/query substrate through electricsheephq/electric-sheep#2026, #2027, and #2028.

This issue remains the injection-mode roadmap anchor. The concrete plugin tool implementation is now tracked in #26:

• Resolve account/workspace first through explicit company_brain_* tools.
• Inject account-scoped context only after deterministic resolution.
• Preserve source citations and insufficient_evidence responses instead of flattening data into generic memory.

Upstream epic: electricsheephq/electric-sheep#2029.

[100yenadmin]

Roadmap hygiene pass after cross-repo audit on May 8, 2026.

Keep this issue as the injection-mode roadmap anchor. Concrete execution is split as:

• Company Brain MCP tools for account resolution and evidence-backed briefings #26 for the actual company_brain_* account-resolution, brief, timeline, and query tools.
• Company Brain plugin canary and operator docs #27 for plugin canary/operator docs.
• Action-safety bridge for Company Brain context #25 for action-safety/approval semantics.

Direction: account context should be injected only after explicit Company Brain account/workspace resolution, and GBrain/shadow context should remain marked non-authoritative where applicable.

[100yenadmin]

Planning alignment update after plugin #28 landed.

Implementation should now build on the explicit company_brain_* tools already in the plugin rather than adding another generic memory path. Suggested PR scope:

• Resolve an account deterministically with company_brain_accounts_list.
• Fetch brief/timeline/query context only for the resolved account.
• Inject a distinct <company-brain-context> block or equivalent labeled section, separate from personal memory and non-authoritative GBrain/shadow context.
• Preserve citations, insufficient_evidence, requires_approval, action_readiness, and visibility_scope.

Dashboard dependencies are now dashboard#157 and dashboard#155; staged validation is electricsheephq/electric-sheep#2040.

[100yenadmin]

Activation slice landed in #29.

The plugin now has an opt-in Company Brain context path that resolves account context through company_brain_accounts_list, fetches account-scoped brief/follow-up context, and injects a distinct Company Brain context block instead of flattening it into generic memory. Citations and insufficient-evidence signals are preserved.

Keeping this issue open as the broader injection-mode roadmap anchor, especially for future non-authoritative GBrain/shadow-context labeling and production runtime refinements.

[100yenadmin]

Sprint update: Company Brain context injection has landed on main, with explicit account-scoped <company-brain-context> injection, account resolution through company_brain_accounts_list, cited brief/timeline context, and no flattening into generic memory. Follow-up PR #30 fixed packaged manifest parity so the context config is discoverable from dist/openclaw.plugin.json as well. Closing as satisfied for the pilot context bridge.