Elicit Brain

Use brain elicit to surface tacit knowledge — things the corpus shows the user knows implicitly (referenced constantly, implied everywhere) but has never written as an authored vault note. The interaction is draft-then-correct: the system makes a confident guess at the unwritten rule and the user corrects it. The correction is the elicited knowledge. Codified notes are tagged tacit + the signal kind and include a ## Source section wikilinking the evidence documents.

For plain document lookup and Q&A, use consult-brain. For themes and patterns across interactions, use brain-graph. For action-item queries, use brain-todo.

When to use this skill

The gap signals

Viewing the gap queue — brain elicit list

Read-only. Runs the active detectors, upserts results, and renders the ranked open gaps. Does not need Ollama.

brain elicit list                    # refresh detectors and show the queue
brain elicit list --json             # machine-readable (for synthesis)
brain elicit list --limit 10         # show at most 10 gaps
brain elicit list --low-confidence   # include gaps below the score floor
brain elicit list --type person --type org   # only person/org gaps (repeatable)

JSON shape per gap: signal_kind, target_type, target_id, score, evidence_ids (list of doc IDs), rationale, status.

--type filters the queue to one or more entity types (person / org / project / topic / tool / doc, repeatable) — pass it when the user scopes the ask ("what haven't I written down about my projects").

Interactive loop — brain elicit

Requires Ollama. For each gap in the ranked queue the command drafts a confident rule text and opens it in $EDITOR. The user corrects/accepts; on a non-empty, changed body the result is codified as a vault note.

brain elicit                                 # run the loop over the full ranked queue
brain elicit --target "engineering culture"  # user-flagged: jump to this specific topic
brain elicit --signal delta                  # only surface delta gaps this run
brain elicit --signal orphan
brain elicit --signal contradiction          # needs BRAIN_ELICIT_CONTRADICTION_ENABLED=true
brain elicit --type project                  # only draft gaps about projects (repeatable)
brain elicit --include-low-confidence        # include below-score-floor gaps

--type (repeatable: person / org / project / topic / tool) narrows the loop to gaps about those entity types — combine with --signal to be precise ("interview me about my orgs, delta signal only").

Interactive keymap:

The editor buffer opens with the drafted rule and a comment block listing the evidence doc IDs. The body must be non-empty and changed from the draft — re-saving the draft unchanged re-prompts rather than codifying.

After codification

A codified gap produces:

• A vault note at brain note new's path, tagged tacit + signal kind.
• A ## Source footer with [[wikilinks]] to the evidence documents.
• The elicitation_gaps row updated to status='resolved' — the gap will not resurface.

Read the new note with brain show <id-prefix> — it is one document you just wrote, so the body is bounded and known. Search future sessions with brain search "<topic>" --tag tacit --json --brief; to read across several tacit notes at once, brain recall "<topic>" --tag tacit --budget 2000 --json keeps the read under a token ceiling. See consult-brain for the full cost-ordered table.

Config knobs

Prerequisites check

Before running brain elicit (the loop), verify:

brain doctor        # Postgres + Ollama must be healthy
brain elicit list   # confirms the table exists and detectors run cleanly

If brain doctor reports Ollama unavailable, brain elicit list still works (Ollama-free). The interactive loop (brain elicit) will fail until Ollama is running.

When NOT to use this skill

• Plain "find docs about X", a quote, a single fact → consult-brain.
• "Themes in my conversations with X", "what connects A and B" → brain-graph.
• "What are my open action items?" → brain-todo.
• The user explicitly says "don't elicit" / "just search my brain".