WL Co-Pilot

Keyboard shortcuts

⌘KOpen search + command palette /Same as ⌘K ?Open this help rRefresh data (poll now + cost-burn + funnel) 1-9, 0Switch tabs (in tab-bar order) ↑↓Navigate search results ↩Open focused search result / fire command escClose any overlay ⌘+↩Fire specialist when in Console prompt

Command palette (⌘K)

Type any word → search ledger, chat, brain, approvals, autopsies
Type "pause" → suggests action approve:pause-both-adsets
Type "resume" → suggests action approve:resume-both-adsets
Type "audit funnel" / "funnel" → suggests analyze:funnel-today
Type "audit ads" → suggests analyze:full-audit
Type "simulate" → suggests simulate-sale (fires Forensic Opus 4.7)
Use the scope chips (all / ledger / chat / brain) to narrow

Specialist Console (tab 8)

ad-optimizer — fatigue + auction + CTR diagnosis, drafts approval if action needed
forensic — deep autopsy via Opus 4.7. Use for real or simulated session_id. ~2-3 min, $1+
compactor — rebuild state.md from ledger. Pure Python, ~3s
analyst — run full analyst_run.sh (current-day KPI synthesis + TG digest)

Approvals grammar (used by approve action / command palette)

approve:pause-ad-<ad_id> · approve:pause-both-adsets · approve:resume-both-adsets
approve:<approval_id> / deny:<approval_id> — drain a queued approval
analyze:ad-<ad_id> · analyze:full-audit · analyze:funnel-today
simulate-sale — spawn Forensic Opus 4.7 on a synthetic session

Export & data flow

System tab → ⬇ buttons download ledger / chat / brain / everything
Brain tab → edit playbook, voice rubric, known_issues, experiments. Saves queue to dispatcher (~20s write-back)
state.md is a projection from ledger.jsonl. If they disagree, ledger wins. Don't hand-edit state.md.

Hard rules (in case Co-Pilot ever feels off-script)