adcp-governance

AdCP Governance Protocol

This skill enables you to execute the AdCP Governance Protocol with governance agents. Covers four areas: property lists (site-level targeting), collection lists (program-level targeting), content standards (brand safety rules), and campaign governance (plans, checks, outcomes, audit trail).

Buyer-side basics — idempotency replay, oneOf variants, async status:'submitted' polling, error recovery from adcp_error.issues[] — live in skills/call-adcp-agent/SKILL.md. This skill covers per-task semantics only.

Overview

The Governance Protocol provides 21 standardized tasks across four areas:

Property Lists

Task	Purpose	Response Time
create_property_list	Create include/exclude list	~1s
update_property_list	Modify list filters/properties	~1s
get_property_list	Retrieve list with optional resolution	~1-5s
list_property_lists	List all accessible lists	~1s
delete_property_list	Delete a list	~1s

Collection Lists

Task	Purpose	Response Time
create_collection_list	Create program-level list	~1s
update_collection_list	Modify list	~1s
get_collection_list	Retrieve with optional resolution	~1-5s
list_collection_lists	List all accessible lists	~1s
delete_collection_list	Delete a list	~1s

Content Standards

Task	Purpose	Response Time
create_content_standards	Create brand safety rules	~1s
get_content_standards	Retrieve standards by ID	~1s
update_content_standards	Modify rules	~1s
list_content_standards	List all accessible standards	~1s
calibrate_content	Test content against standards	~5-30s
get_media_buy_artifacts	Get creatives for compliance review	~5s
validate_content_delivery	Audit delivery compliance	~10-60s

Campaign Governance

Task	Purpose	Response Time
sync_plans	Push or update a campaign plan with budget authority and policies	~1s
check_governance	Validate an action (intent or execution) against the plan	~1-5s
report_plan_outcome	Report a completed action so plan budget state advances	~1s
get_plan_audit_logs	Retrieve governance state, budget tracking, and audit trail	~1-5s

Experimental in 3.0. Campaign governance may change between 3.x releases with at least 6 weeks' notice. Sellers MUST declare governance.campaign in experimental_features to participate. See experimental status.

Typical Workflow

Property Lists (site-level)

1. Create list: create_property_list with base properties and filters
2. Resolve: get_property_list with resolve: true to see matched properties
3. Refine: update_property_list to adjust filters
4. Apply: Reference list_id in create_media_buy targeting

Collection Lists (program-level, CTV)

1. Create list: create_collection_list with distribution IDs or genre filters
2. Resolve: get_collection_list with resolve: true to see matched programs
3. Apply: Reference in campaign targeting for CTV brand safety

Content Standards

1. Create standards: create_content_standards with rules
2. Calibrate: calibrate_content with test samples to validate configuration
3. Monitor: get_media_buy_artifacts + validate_content_delivery for ongoing compliance

Campaign Governance

1. Sync governance agents to seller accounts via sync_governance (lives in the Accounts protocol)
2. Register the plan: sync_plans with budget authority, channel allocation, and resolved policy IDs
3. Validate actions: check_governance on intent (pre-discovery) and execution (pre-buy). Returns an opaque governance_context token the seller echoes on subsequent checks.
4. Report outcomes: report_plan_outcome so committed/remaining budget advances
5. Audit and produce shareable views: get_plan_audit_logs filtered by governance_contexts for the requesting party. See audit trail: internal vs shareable views.

---

Task Reference

create_property_list

Create a property list for brand safety and inventory targeting.

Request:

{
  "name": "Premium News Properties",
  "description": "Tier 1 news publishers for brand campaigns",
  "base_properties": [
    {
      "selection_type": "publisher_tags",
      "publisher_domain": "publisher.com",
      "tags": ["premium", "news"]
    }
  ],
  "filters": {
    "countries_all": ["US", "GB"],
    "channels_any": ["display", "video"]
  },
  "brand": {
    "domain": "acmecorp.com"
  }
}

Key fields:

• name (string, required): Human-readable name
• description (string, optional): Purpose of the list
• base_properties (array, optional): Property sources — publisher_tags, publisher_ids, or identifiers
• filters (object, optional): Resolution filters — countries_all, channels_any, property_types, feature_requirements, exclude_identifiers
• brand (object, optional): Brand reference for automatic rule inference

---

update_property_list

Modify an existing property list.

Request:

{
  "list_id": "pl_abc123",
  "filters": {
    "countries_all": ["US", "GB", "DE"],
    "channels_any": ["display", "video", "ctv"]
  }
}

Key fields:

• list_id (string, required): Property list identifier
• name, description (string, optional): Update metadata
• base_properties (array, optional): Replace property sources
• filters (object, optional): Replace filter configuration

---

get_property_list

Retrieve a property list with optional resolution.

Request:

{
  "list_id": "pl_abc123",
  "resolve": true,
  "max_results": 50
}

Key fields:

• list_id (string, required): Property list identifier
• resolve (boolean, optional): Resolve filters and return property identifiers (default: false)
• max_results (number, optional): Max properties when resolved

---

list_property_lists

List all property lists accessible to the authenticated principal.

Request:

{
  "name_contains": "premium"
}

Key fields:

• name_contains (string, optional): Filter by name substring
• max_results (number, optional): Max results

---

delete_property_list

Delete a property list.

Request:

{
  "list_id": "pl_abc123"
}

Key fields:

• list_id (string, required): Property list identifier to delete

---

create_collection_list

Create a collection list for program-level brand safety (CTV, podcast, streaming).

Request:

{
  "name": "Family-Safe CTV Programs",
  "description": "Programs suitable for family brand campaigns",
  "base_collections": [
    {
      "selection_type": "publisher_genres",
      "publisher_domain": "ctv-publisher.com",
      "genres": ["family", "comedy"],
      "genre_taxonomy": "iab_content_taxonomy_3.0"
    }
  ],
  "filters": {
    "content_ratings_exclude": [
      { "system": "us_tv", "rating": "TV-MA" }
    ],
    "kinds": ["series"]
  },
  "brand": {
    "domain": "familybrand.com"
  }
}

Key fields:

• name (string, required): Human-readable name
• base_collections (array, optional): Collection sources — distribution_ids, publisher_collections, or publisher_genres
• filters (object, optional): content_ratings_exclude, content_ratings_include, genres_exclude, genres_include, kinds, production_quality
• brand (object, optional): Brand reference

Distribution identifier types: imdb_id, gracenote_id, eidr_id

---

update_collection_list

Modify an existing collection list.

Request:

{
  "list_id": "cl_abc123",
  "filters": {
    "content_ratings_exclude": [
      { "system": "us_tv", "rating": "TV-MA" },
      { "system": "us_tv", "rating": "TV-14" }
    ]
  }
}

Key fields:

• list_id (string, required): Collection list identifier
• base_collections, filters (optional): Replace configuration

---

get_collection_list

Retrieve a collection list with optional resolution.

Request:

{
  "list_id": "cl_abc123",
  "resolve": true
}

Key fields:

• list_id (string, required): Collection list identifier
• resolve (boolean, optional): Resolve and return collection entries
• max_results (number, optional): Max collections when resolved

---

list_collection_lists

List all collection lists accessible to the authenticated principal.

Request:

{
  "name_contains": "family"
}

---

delete_collection_list

Delete a collection list.

Request:

{
  "list_id": "cl_abc123"
}

---

create_content_standards

Create content standards (brand safety rules) for campaign compliance.

Request:

{
  "name": "Automotive Brand Safety",
  "description": "Content rules for automotive brand campaigns",
  "rules": [
    { "rule_type": "category", "action": "block", "value": "violence", "severity": "critical" },
    { "rule_type": "category", "action": "block", "value": "adult", "severity": "critical" },
    { "rule_type": "keyword", "action": "flag", "value": "accident", "severity": "medium" }
  ],
  "brand": {
    "domain": "automaker.com"
  }
}

Key fields:

• name (string, required): Human-readable name
• rules (array, optional): Content rules — rule_type, action (allow/block/flag), value, severity
• brand (object, optional): Brand reference for automatic rule inference

---

get_content_standards

Retrieve content standards by ID.

Request:

{
  "standards_id": "cs_abc123"
}

---

update_content_standards

Modify existing content standards.

Request:

{
  "standards_id": "cs_abc123",
  "rules": [
    { "rule_type": "category", "action": "block", "value": "violence", "severity": "critical" }
  ]
}

---

list_content_standards

List all content standards accessible to the authenticated principal.

Request:

{
  "name_contains": "automotive"
}

---

calibrate_content

Test content samples against content standards to validate configuration.

Request:

{
  "standards_id": "cs_abc123",
  "samples": [
    { "url": "https://example.com/article1", "expected_result": "allow" },
    { "url": "https://example.com/article2", "expected_result": "block" },
    { "text": "Car crash injures three people", "expected_result": "block" }
  ]
}

Key fields:

• standards_id (string, required): Content standards to calibrate against
• samples (array, required): Content samples with url and/or text, and optional expected_result

---

get_media_buy_artifacts

Get creative artifacts from a media buy for compliance review.

Request:

{
  "media_buy_id": "mb_abc123",
  "sales_agent_url": "https://sales.publisher.com"
}

Key fields:

• media_buy_id (string, required): Media buy identifier
• sales_agent_url (string, required): Sales agent that owns the media buy

---

validate_content_delivery

Validate delivered content against content standards.

Request:

{
  "standards_id": "cs_abc123",
  "media_buy_id": "mb_abc123",
  "sales_agent_url": "https://sales.publisher.com",
  "date_range": {
    "start": "2025-01-01",
    "end": "2025-01-31"
  }
}

Key fields:

• standards_id (string, required): Content standards to validate against
• media_buy_id (string, required): Media buy identifier
• sales_agent_url (string, required): Sales agent URL
• date_range (object, optional): Filter by delivery date range

---

sync_plans

Register or update a campaign plan that defines authorized parameters (budget, channels, policies) for an orchestrator's autonomous action.

Request:

{
  "plan_id": "plan_q1_2026_launch",
  "plan_version": 1,
  "budget": { "authorized": 500000, "currency": "USD" },
  "channel_allocation": { "olv": 0.55, "display": 0.30, "audio": 0.15 },
  "policies": ["us_coppa", "alcohol_advertising"],
  "human_review_required": false
}

Key fields:

• plan_id (string, required): Stable plan identifier
• plan_version (integer, required): Increment on every modification — checks bind to a version
• budget.authorized (number, required): Total spend authority across all governed actions on this plan
• policies (array, optional): Registry policy IDs that govern this plan. Inline custom_policies may add restrictions but cannot relax registry policies.
• human_review_required (boolean, optional): Force human review on all actions. Auto-set true when any resolved policy has requires_human_review: true.

---

check_governance

Validate an action against the plan. Called twice in the lifecycle: once on intent (pre-discovery), once on execution (pre-buy). Sellers MUST call execution checks independently using credentials synced via sync_governance.

Request (execution check):

{
  "plan_id": "plan_q1_2026_launch",
  "plan_version": 1,
  "purchase_type": "media_buy",
  "tool": "create_media_buy",
  "payload": { "...": "the create_media_buy request body" },
  "governance_context": "gc_mb_seller_456"
}

Key fields:

• purchase_type (enum, required): media_buy, rights_license, signal_activation, or creative_services
• governance_context (string, optional): Echoed on subsequent checks for the same governed action. The agent issues this on the first check and the buyer attaches it to the action envelope.
• tool (string, required): Which AdCP tool is being authorized
• payload (object, required): The full request body the orchestrator/seller would otherwise send

Response status: approved, denied, or conditions. On denied, read governance_context.findings[] to locate the failed rule and correct the payload.

---

report_plan_outcome

Report the result of a governed action so plan budget and state advance.

Request:

{
  "plan_id": "plan_q1_2026_launch",
  "governance_context": "gc_mb_seller_456",
  "outcome": "completed",
  "committed_budget": 150000
}

Key fields:

• governance_context (string, required): The token issued on the original check
• outcome (enum, required): completed, cancelled, failed, or delivery (for ongoing pacing reports)
• committed_budget (number, required for completed): Net budget committed by this action

---

get_plan_audit_logs

Retrieve governance state, budget tracking, and audit trail for one or more plans.

Request:

{
  "plan_ids": ["plan_q1_2026_launch"],
  "governance_contexts": ["gc_mb_seller_456"],
  "include_entries": true
}

Key fields:

• plan_ids / portfolio_plan_ids / governance_contexts (at least one required): Scope the query
• include_entries (boolean, optional): Return the full audit trail. Default false returns summary only.

Producing a shareable view: filter governance_contexts to the requesting party's actions and strip plan-level aggregates (budget.*, channel_allocation.*, summary.drift_metrics) before forwarding. See audit trail: internal vs shareable views.

---

Key Concepts

Property Lists vs Collection Lists

• Property Lists: Site-level targeting. Operates on publisher domains and properties (websites, apps, CTV apps). Use for "where" the ad appears.
• Collection Lists: Program-level targeting. Operates on shows, series, and content programs using distribution identifiers (IMDb, Gracenote, EIDR). Use for "what content" the ad appears alongside (primarily CTV).

Content Standards vs Property/Collection Lists

• Content Standards: Rules-based evaluation of content quality, topics, and safety. Evaluates content dynamically.
• Property/Collection Lists: Pre-computed sets of approved or excluded inventory. Static targeting applied at campaign setup.

Filter Resolution

Property and collection lists combine static selections with dynamic filters. Use resolve: true on get operations to see the final resolved set of properties or collections.

Three invariants for audit and disclosure decisions

These three properties of campaign governance shape what an orchestrator can disclose, can rely on a counterparty having, and cannot work around. Surface them when audit-trail design or counterparty disclosure decisions come up.

1. Inline policies are additive-only over registry policies. A buyer's bespoke custom_policies (or inline policy entries on a plan) may add restrictions on top of registry-sourced policies. They MUST NOT relax, override, or disable registry policies. Counterparties who see policies_evaluated: ["us_coppa"] can trust the registry version of us_coppa was applied at its declared enforcement level.
2. governance_context is the seller-visible correlation token; full plan/budget data is buyer-side. The seller sees the opaque token they were issued and the entries scoped to it. Plan-level totals (budget.authorized, channel_allocation, drift_metrics) belong to the buyer's internal view and are never shared by default.
3. plan_hash is the cryptographic attestation surface. base64url_no_pad(SHA-256(JCS(plan_payload))) over the plan revision the check evaluated. Any party with the plan revision can recompute and byte-compare. This is what makes a four-field shareable attestation (governance_context, status, plan_hash, policies_evaluated) cryptographically meaningful — counterparties don't have to trust the buyer's summary.

A related working-group adoption pattern — effective_date enabling informational-before-enforcement of new policies — lives in Policy Registry; it shapes registry rollout rather than per-check disclosure decisions.

---

Error Handling

Common error codes:

• LIST_NOT_FOUND: Invalid list_id
• STANDARDS_NOT_FOUND: Invalid standards_id
• UNAUTHORIZED: Not authorized to access this resource
• VALIDATION_ERROR: Invalid filter or rule configuration
• PLAN_NOT_FOUND: No plan with this ID, or the principal is not authorized for it. Returned indistinguishably from the unauthorized case to prevent plan-ID enumeration.
• GOVERNANCE_DENIED: check_governance rejected the action. Read governance_context.findings[] to identify the failed rule, correct the payload, and retry.