The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a local Node.js script (scripts/arms.mjs) to perform statistical comparisons and test planning. It also provides instructions for using ffmpeg to extract frames from video assets for creative review. These executions are scoped to local scripts and standard system utilities provided or expected within the project environment.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external ad platforms (Meta Ads Manager) and backend database records (leads and payments), which presents a theoretical surface for indirect prompt injection.
Ingestion points: Data enters the agent's context through ad performance reports, campaign metadata, and backend reconciliation tables.
Boundary markers: The skill explicitly enforces 'Hard gates' in SKILL.md, stating that nothing delivers an impression without the user's explicit consent and that all created objects must remain in a PAUSED status until verified.
Capability inventory: The skill has the capability to write results to local files and execute CLI tools for math and media processing.
Sanitization: The documentation includes detailed procedures for identifying and filtering suspected automated sessions (crawlers) and emphasizes triangulating data across multiple sources to ensure accuracy.

ad-test-loop