kmg-auto-recall

name	kmg-auto-recall
description	Auto-invoke knowledge graph search when user asks about project history, past decisions, or previously solved problems

Skill: kmg-auto-recall

Purpose: Auto-invoke knowledge graph search when user asks about project history, past decisions, or previously solved problems.

Trigger Keywords:

• "have we done this before"
• "what did we decide about"
• "did we figure out"
• "is there a lesson on"
• "what do we know about"
• "have I seen this before"

Note on planning contexts: In brainstorming and planning workflows, recall is enforced via HARD BLOCK injection in pre-skill-rules-inject.sh — not via these trigger keywords. This skill handles reactive ("have we?") recall only.

Behavior: When triggered:

1. Extract the user's search query/keywords from their question
2. Invoke the kmgraph:kmg-recall command via the Skill tool with the extracted query
   • The command chains through gov-capture-routing (level/scope detection) → recall-agent (search)
   • This is NOT a direct recall-agent dispatch; level routing happens first
   • gov-capture-routing is Claude Code-only. On non-Claude platforms or if unavailable: dispatch directly to recall-agent with --scope=active as fallback (UQ-8)
3. Present results naturally — do not expose internal routing mechanics

Priority enforcement: Recall results take priority — the model must reason about findings before recommending:

• If recall surfaces a rejected approach, examine WHY it was rejected and whether that reason is still applicable.
• If still applicable: do not propose the approach; if unavoidable, explain why no workaround exists.
• If no longer applicable: may propose it, but must document why the old rejection no longer holds AND lay out the full cascade impact.
• If it is the only viable option: propose it, but lay out complete ramifications and cascade effects across all affected systems.
• If nothing found: state "No prior art found for [topic]." and continue.

Example Trigger:

User: "Have we solved database migration issues before?
I'm seeing timeout errors on large tables."

Assistant Response: "Let me check what we've documented about this before answering..." → [Invoke kmgraph:kmg-recall via Skill tool with query: "database migration timeout errors"] → [Command routes through gov-capture-routing → recall-agent] → [Surface relevant lessons/decisions, including related patterns]