suggest-connections

SKILL: Brain Connection Proposals

This skill generates evidence-backed, read-only connection proposals for the configured Brain and exposes the same queue to the inbox and global graph.

ID: GRAPH-MGMT-20260408 Status: ACTIVE

---

1. Canonical generation

Analyze Brain reading and manual permanent notes for connections, support, contradictions, and gaps.

• Config: <vault>/.gnosi/llm_wiki.json and the managed Brain agent.
• Source of truth: <vault>/.gnosi/llm_wiki_suggestions.json.
• CLI: python3 -c "from backend.services.llm_wiki_actions import run_maintenance; print(run_maintenance(semantic=True))"
• Runtime: the suggest_connections scheduler invokes the same service.

---

2. Review contract

• Proposals may be inspected, opened, or dismissed.
• The automation never creates or edits a permanent note.
• Dismissal updates the canonical queue and invalidates the graph response.
• Only existing Brain page IDs may become proposal members.

---

3. Model and scheduling

• Model selection follows the managed Brain agent and Gnosi's provider factory.
• Manual semantic maintenance and the explicitly enabled suggest_connections scheduler may invoke the model.
• Daily LLM Wiki maintenance and update_memories are deterministic and must never invoke semantic generation.

---

4. Restrictions and learned safeguards

• Do not write a second generated graph under BD/; it diverges from the inbox.
• Do not call the removed suggest_connections_digital_brain module.
• Do not report a structured generator error as scheduler success.
• Generating proposals can read selected vault content and invoke configured model providers. Run the CLI only for requested generation; use synthetic proposals and fake providers for tests, never a real vault as incidental QA.
• Add proposal edges directly from the queue as a non-structural response overlay, and invalidate the graph response cache whenever the queue changes.