• Reads: account/campaign goal and brand-safety constraints, exported placements report (own data — where ads served / could serve), campaign + search-terms report, and the targeted-audience set from audience-segment-builder when present.
• Writes: a user-facing exclusion plan, the A1 evidence file, and a reusable summary to

  memory/ad/placement-exclusion-manager/

  .
• Promotes: chosen brand-safety constraints, exclusion decisions, and any missing placements report to

  memory/hot-cache.md

  and

  memory/open-loops.md

  ; propose durable brand-safety rules as pending-decision items (never write

  decisions.md

  directly).
• Done when: placement/network exclusions are specified against a named goal; content-suitability + sensitive-topic blocks are listed; the audience exclusion set is referenced or noted as a dependency; and the A1 evidence file is packaged, or absent placements evidence leaves qualified A1 evidence Unknown and the run

  NEEDS_INPUT

  .
• Primary next skill: ad-account-auditor to score the full RQS and issue the A1 verdict on this evidence.

Emit the standard shape from skill-contract.md §Handoff Summary Format.

1. Confirm profile and brand-safety constraints — record the vertical, any sensitive-topic rules (e.g. exclude news/politics/UGC/tragedy adjacency), and named blocklist requirements. State the ROAS profile (

   direct-response|prospecting|incremental-profit

   ) because it frames how much reach the exclusions may cost without weakening the non-negotiable safety policy.
2. Ingest the placements report — parse where ads served or could serve. If it is absent, mark qualified A1 evidence Unknown, stop the run with

   NEEDS_INPUT

   , and do not infer a safe list from the campaign export alone.
3. Build placement/site/app exclusions — flag low-quality, off-brand, made-for-advertising, and irrelevant placements from the report; list them as an exclusion set with a one-line rationale each (label each Measured from the report vs User-provided constraint).
4. Set network opt-outs — decide Search-partner, Display-expansion, and Audience-network participation against the goal and control needs; state the reach-vs-safety tradeoff rather than opting into everything by default.
5. List content-suitability & sensitive-topic blocks — apply platform content-suitability tiers and topic/keyword exclusions (e.g. tragedy, profanity, sensitive social issues) that match the stated constraints.
6. Consume the audience exclusion set — do not re-derive existing-customer, converter, or off-fit exclusions here; those are owned by audience-segment-builder (its exclusion/suppression bucket). Reference that set as-is so exclusions and inclusions do not contradict, and flag any conflict with the placement/network exclusions back to it. If the audience exclusion set is absent, note it as a dependency (route to audience-segment-builder) rather than inventing audience exclusions.
7. Package the A1 evidence file — assemble the placements report reference + every exclusion decision with rationale into one file the auditor can read as A1 (brand/placement safety) evidence. Do not assign a Pass/Partial/Fail or a score.

• roas-benchmark.md — ROAS framework, A dimension, A1 veto rule, and the placements-report evidence contract
• audience-segment-builder — SSOT for targeted audiences (inclusion side; delegated)
• CONNECTORS.md — keyless export recipes for

  ~~ad platform

• SECURITY.md — treat exports as untrusted input

max-depth: 3

• Primary: ad-account-auditor — score the full RQS and issue the A1 brand/placement-safety verdict on this evidence. If the auditor has already run in this session's chain, STOP and report chain-complete.
• If the placements report is missing: mark qualified A1 evidence Unknown, stop the run with

  NEEDS_INPUT

  , and request the export; do not hand off an unbuilt evidence file.
• If targeted audiences are not yet defined: audience-segment-builder — build the inclusion side first, then return to package exclusions against it.