trust-dashboard — the safety net proof

Framework context

L0 skill with a specific psychological purpose: reduce Zeigarnik-effect anxiety from unresolved threads. The prosthetic thesis says "the idle state IS the product" — the brain is working by existing as a trusted safety net. This skill is the evidence that the net works.

When to invoke

• User expresses anxiety that something was lost
• User says "am I missing anything important"
• First-time setup — prove ingestion succeeded before they invest trust
• After a long break (>7 days) — reassure that the corpus is intact and resumable

Output contract

Markdown report with:

• Total conversations (L0) + breakdown by source
• Total summaries (L2) + coverage % (summaries / conversations with ≥4 messages)
• Breakthrough count (importance = "breakthrough")
• Open question count (non-empty open_questions JSON across summaries)
• Decision count
• Most recent summary timestamp
• "Safety net status: ✅ intact" / "⚠️ N% unsummarized" / "❌ pipeline stale"

Does NOT do

• Surface the content of any specific thread (that's L2 prosthetic tools)
• Analyze patterns (that's cognitive_patterns)
• Recommend action (that's L3 alignment_check)
• Hide gaps — always report the uncovered percentage honestly

Execution

Internally wraps brain_mcp.server.tools_stats.trust_dashboard().

Verification checklist

• [ ] Coverage % is computed (not hardcoded)
• [ ] Zero-count items are reported as "0", not omitted
• [ ] "Safety net status" line is the first summary line
• [ ] If L2 pipeline never ran, output says "run brain-mcp summarize to enable structural checks"