The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from ad platform APIs which could contain malicious instructions.
Ingestion points: Processes google.json (search terms report containing user-generated queries) and meta.json/linkedin.json (creative details including headlines and descriptions).
Boundary markers: No explicit delimiters or instructions to ignore embedded content are used when processing platform data.
Capability inventory: The skill can execute local Python scripts (pull_all.py, meta.py, etc.) and write files to the workspace/ directory.
Sanitization: No evidence of sanitization or filtering of external text inputs before they are incorporated into the review process.
[COMMAND_EXECUTION]: The skill executes locally stored Python scripts to perform ad management tasks.
Runs scripts/ads/pull_all.py to retrieve performance metrics.
Mentions mutation capabilities (pausing ads, changing budgets) via meta.py, google.py, and linkedin.py scripts.

kai-daily-ad-review