The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests external data from the Meta Marketing API, specifically ad headlines and body text via scripts/get_active_ads_copy.py. This content is processed by the agent, creating a surface where malicious instructions inside ad creative could attempt to influence the agent's logic.
Ingestion points: Ad copy fetched from the Graph API in scripts/get_active_ads_copy.py.
Boundary markers: Ad content is processed without specific delimiters or instructions to ignore embedded commands.
Capability inventory: The skill can create or modify campaigns and upload custom audiences (scripts/create_campaign.py, scripts/update_campaign.py, scripts/create_custom_audience.py).
Sanitization: No sanitization or filtering of fetched ad text is performed before display or interpretation.
[EXTERNAL_DOWNLOADS]: The skill relies on standard, well-known Python packages and contains references to example URLs.
Evidence: The skill requires requests, python-dotenv, and tabulate, which are standard for API interaction and data formatting.
URL Findings: Automated scans flagged landing.com as a malicious URL. This domain appears as a hardcoded placeholder in scripts/get_active_ads_copy.py and as an example in api-reference.md. In this context, it is used strictly as a template value and does not facilitate an active attack.
[COMMAND_EXECUTION]: The skill includes several scripts that interact with the Meta Graph API to perform account management tasks.
Evidence: Scripts such as create_campaign.py, ad_scheduler.py, and update_campaign.py execute state-changing operations on the Meta Ads platform. These operations are the intended primary purpose of the skill and require valid user credentials provided via the environment.

meta-ads