wicked-brain-enhance

name	wicked-brain-enhance
description	Fill gaps in brain knowledge. Dispatches an enhance subagent that identifies thin areas and writes inferred chunks to expand coverage. Use when: "enhance the brain", "fill gaps", "brain enhance", "what's missing in the brain".

wicked-brain:enhance

Enhances the brain by dispatching a subagent that fills knowledge gaps.

Cross-Platform Notes

Commands in this skill work on macOS, Linux, and Windows. When a command has platform differences, alternatives are shown. Your native tools (Read, Write, Grep, Glob) work everywhere — prefer them over shell commands when possible.

For the brain path default:

• macOS/Linux: ~/.wicked-brain
• Windows: %USERPROFILE%.wicked-brain

Config

Brain discovery + server lifecycle are handled by wicked-brain-call. Pass --brain <path> to override the auto-detected brain, or set WICKED_BRAIN_PATH. The CLI starts the server on first call (no manual init required) and writes an audit record to {brain}/calls/ per call.

Process

Dispatch an enhance subagent with these instructions:

You are a knowledge enhancement agent for the digital brain at {brain_path}.
Server interactions: use `npx wicked-brain-call <action> [--param k=v ...]`.

## Step 1: Find gaps

Read the recent event log using your Read tool on `{brain_path}/_meta/log.jsonl`
(read the last 100 lines). Shell fallback:
- macOS/Linux: `tail -100 {brain_path}/_meta/log.jsonl`
- Windows: `Get-Content "{brain_path}\_meta\log.jsonl" -Tail 100`

Get stats:
```bash
npx wicked-brain-call stats

Read _meta/log.jsonl. Look for:

1. op: "search_miss" entries — these are topics users searched for but couldn't find. Prioritize these as highest-value gaps.
2. Low-frequency tags in chunks (existing approach) — secondary signal.

Group search misses by topic and address the most frequent ones first.

Search for thin areas — topics mentioned in existing chunks but with few entries. Use your Grep tool on {brain_path}/chunks/ to find all contains: fields and count occurrences. Shell fallback:

• macOS/Linux: grep -roh 'contains:' {brain_path}/chunks/ -A 5 2>/dev/null | grep ' - ' | sort | uniq -c | sort -n
• Windows: Select-String -Recurse -Pattern " - " "{brain_path}\chunks\*.md" 2>nul | Select-Object -ExpandProperty Line | Sort-Object | Group-Object | Sort-Object Count

Step 2: Identify what's missing

Based on existing content, reason about:

• Topics mentioned but never elaborated
• Connections between concepts that exist but aren't documented
• Questions the brain can't currently answer

Source Material Rules (CRITICAL)

ONLY read from chunks/extracted/ as source material for new inferences. Never base inferences on content from chunks/inferred/ — that would be inference-of-inference, which causes confidence laundering (inferred content cites inferred content, making unreliable chains appear well-sourced).

Why this matters: Each inference step introduces uncertainty. When you infer from already-inferred content, errors and assumptions compound silently across the chain. The result is chunks that sound authoritative but are actually several degrees removed from any real source. A confidence: 0.6 chunk derived from another confidence: 0.6 chunk is not confidence: 0.6 — it is substantially less reliable, but the metadata won't show it. Only raw source material (extracted chunks, actual files) provides a trustworthy evidence base.

• If you find relevant content in chunks/inferred/, you may note it as background context but do NOT cite it as a source_chunk or use it as evidence for new inferences.
• Every entry in source_chunks in your output MUST start with chunks/extracted/.
• If a gap cannot be filled using only extracted chunks as evidence, skip it.

Bootstrapping caveat: If the brain contains no chunks/extracted/ files at all (e.g., very early stage with only inferred chunks), do NOT proceed silently. Stop and inform the user: "The brain has no extracted source chunks — enhancement requires raw source material. Run wicked-brain:ingest to add source files first." This prevents a silent chain of inferences with no grounding.

Step 3: Write inferred chunks

For each gap, write a new chunk to {brain_path}/chunks/inferred/{topic}/chunk-NNN.md:

---
source: inferred
source_type: llm
chunk_id: inferred/{topic}/chunk-NNN
content_type:
  - text
contains:
  - {topic tags}
entities:
  systems: []
  people: []
  programs: []
  metrics: []
confidence: 0.6
indexed_at: {ISO timestamp}
authored_by: llm
narrative_theme: {what this fills}
source_chunks:
  - {existing chunk that informed this inference}
---

{Synthesized content based on existing brain knowledge. Do not fabricate facts.
Only synthesize connections and summaries from what already exists.}

Step 4: Index and log

Index each new chunk via the server API. Append to log.jsonl for each chunk written.

Step 5: Emit bus events

For each inferred chunk created, emit:

npx wicked-bus emit \
  --type "wicked.brain.chunk.enhanced" \
  --domain "wicked-brain" \
  --subdomain "brain.chunk" \
  --payload '{"path":"{chunk_path}","brain_id":"{brain_id}","topic":"{topic}","confidence":0.6}' 2>/dev/null || true

Fire-and-forget — if the bus is not installed, silently skip.

Step 6: Report

State what gaps were identified and how many inferred chunks were created.