/ad

Paid Command

Run the paid-ads lifecycle along the ROAS loop (Research → Orchestrate → Activate → Scale). Skills score on the ROAS framework and operate from the user's own-account manual export — keyed ad-platform APIs are never required.

Route

Infer the ROAS-loop phase from the goal (or honor --phase) and route to the matching skill:

• Research — campaign-architect (account/campaign structure), audience-segment-builder (segments), search-term-miner (converting-query mining + negative-keyword lists), product-feed-optimizer (Shopping/PMax feed, title/attribute fixes); reuse budget-optimizer for spend; consult offer-claims-registry's live-offers table (memory/claims/offers.md) when structuring promo campaigns
• Orchestrate — ad-creative-builder, ad-test-designer, bid-strategy-planner (tCPA/tROAS target + learning-phase entry), landing-experience-checker (pre-launch Quality-Score + ad-to-page message-match preflight); reuse landing-optimizer for the post-click page; read approved wording from the claims projection and submit [needs source] items as operation: propose events to the claims registry
• Activate — conversion-signal-qa (build/verify tracking), placement-exclusion-manager (brand-safety exclusion lists), conversion-value-mapper (margin→value so tROAS optimizes profit, not orders), then ad-account-auditor (the RQS gate + launch go/no-go; O1/O2 judged against offer-claims-registry's ledger)
• Scale — paid-measurement-loop, attribution-reconciler, budget-pacing-monitor (over/under-spend pacing), fatigue-frequency-manager (frequency/CTR decay + creative rotation); reuse roi-calculator / report-generator / performance-analyzer

Rules

• ad-account-auditor is the launch gate: score RQS and enforce the five vetoes (R1/R2/O1/O2/A1) before any budget increase; run conversion-signal-qa first so R1/R2 can be trusted, and resolve unregistered claims via offer-claims-registry before the O1/O2 checks.
• memory/events/claims.ndjson is the claims/offers history. Other skills submit authorized operation: propose events through registry-events.py; offer-claims-registry alone accepts/rejects or mutates canonical state, while Markdown ledgers are generated views.
• Keyless Tier 1 — score from native ad-manager / GA4 / ecommerce exports the user provides; keyed Google Ads / Meta APIs are opt-in Tier-2/3 MCP only.
• Only ad-account-auditor computes the profile-weighted RQS; every other skill works one lever and hands off. budget-optimizer (reused from influencer) allocates spend across campaigns; in-platform bid strategy, search-term mining, and feed work are their own skills (bid-strategy-planner, search-term-miner, product-feed-optimizer).
• Label every metric Measured / User-provided / Estimated; never invent ROAS/CPA figures.

Output

Return inline artifacts by default. Files may be written only when the user explicitly asks and the runtime can write.