The Agent Skills Directory

[SAFE]: The skill provides technical guidance for integrating with the Ad Context Protocol. It focuses on protocol invariants such as idempotency keys, async task polling, and structured error recovery.
[EXTERNAL_DOWNLOADS]: References to external protocol documentation and schemas are directed to adcontextprotocol.org and github.com/adcontextprotocol, which are official and authoritative domains for the protocol.
[COMMAND_EXECUTION]: Mentions of commands like npm run sync-schemas are context-specific to SDK installation and setup by a developer, rather than instructions for the agent to perform unauthorized execution.
[PROMPT_INJECTION]: The skill defines a surface for processing data from external agents. While this represents a potential indirect prompt injection vector, the skill's emphasis on JSON schema validation and structured error envelopes provides a significant layer of technical verification for incoming data.
Ingestion points: Data returned from tools get_products, get_adcp_capabilities, get_schema, and tasks/get (SKILL.md).
Boundary markers: Absent.
Capability inventory: Access to mutating tools including create_media_buy, sync_creatives, and activate_signal (SKILL.md).
Sanitization: Relies on protocol-level JSON schema enforcement.

call-adcp-agent