Khaos-Brain: Skill for Codex

KB Organization Contribute

Run one organization contribution pass for this predictive KB repository.

The organization KB is a shared exchange layer. Export only reusable material that other local KBs may use directly from a synchronized snapshot; organization acceptance does not override each machine's local retrieval weighting or Sleep judgment.

Contribution writes only to the incoming lane under kb/imports//. It must never write directly to kb/main; organization maintenance is responsible for reviewing imports and moving accepted material into the main exchange surface. Local download/search reads organization cards from kb/main, not from kb/imports.

An organization card is an exchange projection, not local semantic authority. Contribution derives it only from a current exact local LogicGuard binding and must not export projection-only related_cards as relationship authority. A receiving machine uses the source-qualified card directly; only outcome evidence may later cause its local Sleep to change local weight or derive a distinct local candidate. Contribution never writes a receiving machine's model or mesh.

Authority

Work from the repository root. Treat these files as authoritative before stateful contribution work:

• PROJECT_SPEC.md
• docs/organization_mode_plan.md
• .agents/skills/local-kb-retrieve/SKILL.md

Current user instructions still override repository files.

Execution Contract

1. Use scripts/kb_org_outbox.py --automation as the entry point.
2. The entry point must first read .local/khaos_brain_desktop_settings.json.
3. If organization mode is not connected to a validated organization repository, exit successfully with a no-op result.
4. Sync the validated organization mirror first so contribution compares against current organization main cards and imports before upload. Any retired layout must block this runtime and be handled by the versioned upgrader.
5. Run KB preflight against system/knowledge-library/organization before exporting any proposals.
6. Export only shareable model or heuristic projections whose exact local LogicGuard binding is current, public, privacy-safe, and useful at organization scope.
7. Do not export private cards, personal preferences, credentials, raw local paths, or raw machine identifiers.
8. Use content hashes for duplicate prevention across exported/uploaded local proposals, current organization catalog cards, imports, and prior contribution receipts. Viewed/selected/used interactions are evidence events, not copies or download/adoption states.
9. Put eligible local cards into the organization outbox, then automatically prepare and push an organization import branch under kb/imports// when proposals were created.
10. After branch materialization and immediately before push, revalidate the exact changed files, card counts, content hashes, privacy checkpoint, shareability, Skill author/version/hash metadata, and rollback/base-branch information. A preflight result from before materialization cannot authorize the push.
11. After a successful revalidation and push, open the organization PR when the repository is on GitHub; apply the org-kb:auto-merge label only when the exact changed files and current checks allow it. On validation, push, PR, or label failure, report the concrete terminal and restore the local mirror rather than claiming a partial contribution.
12. Leave movement into organization main, trust upgrades, merge approval, and final organization exchange decisions to organization maintenance and GitHub checks.
13. When a card depends on a local Skill, upload it as a card-bound Skill bundle with bundle_id, content_hash, version_time, original_author, readonly_when_imported: true, and update_policy: original_author_only.
14. If several local cards point at the same bundle_id, upload the local latest version for that bundle, not an older card-carried copy.
15. Include Skill dependencies only when card evidence explains when the Skill is useful, what outcome it predicts, and the current unavailable_skill_guidance; otherwise omit the Skill bundle while preserving the card itself without pretending the Skill is available.
16. Run KB postflight after a non-skipped contribution pass and record the result as structured history.