verify

Knowledge Verify

Independent verification of self-learn knowledge. Runs 50 fresh questions against NotebookLM to catch errors the original validation missed. Verify is separate from self-learn — it owns the audited flag entirely.

Configuration

obsidian_vault: "{vault}/knowledge/raw"  # Read {vault} from ${CLAUDE_PLUGIN_ROOT}/brain-os.config.md
notebooklm_bin: ~/.local/bin/notebooklm

Audit Flags

Flag	/think	/absorb
true (verified)	No warning	Auto-absorb
false (unverified)	Warning	Blocked
manual (re-review)	Warning	Approval prompt

Stored in {book_vault}/_validation/audit-flag.json.

Commands

Always run the script — it handles fuzzy matching and execution.

PLUGIN_ROOT="${CLAUDE_PLUGIN_ROOT:-$(ls -d ~/.claude/plugins/cache/brain-os-marketplace/brain-os/*/ 2>/dev/null | sort -V | tail -1)}"; PLUGIN_ROOT="${PLUGIN_ROOT%/}"
# Run verify
python3 ${PLUGIN_ROOT}/skills/verify/scripts/verify.py {vault}/knowledge/raw <optional-fuzzy-name>

# Check status
python3 ${PLUGIN_ROOT}/skills/verify/scripts/verify.py {vault}/knowledge/raw --status

# Set flag
python3 ${PLUGIN_ROOT}/skills/verify/scripts/verify.py {vault}/knowledge/raw --set-flag <true|false|manual> <optional-fuzzy-name>

How Verify Runs

1. Fuzzy match book name against knowledge/raw/ folders
2. Generate 50 fresh questions (25 topic, 15 cross-cutting, 10 adversarial)
3. Agent answers from Obsidian notes only
4. NotebookLM answers via notebooklm ask
5. LLM-as-judge scores at ≥95 threshold
6. 100% pass → audited: true. Any fail → audited: false, report to inbox

Results saved to {book_vault}/_validation/audit-results-{date}.jsonl.

Outcome log

Follow skill-spec.md § 11. Append to {vault}/daily/skill-outcomes/verify.log:

{date} | verify | verify | ~/work/brain-os-plugin | knowledge/raw/{slug}/_validation/audit-results-{date}.jsonl | commit:{hash} | {result}

• result: pass if 100% questions score ≥95 (audited=true), fail if any question fails (audited=false)
• Optional: args="{book-name}", score={passed}/{total}