The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes several local helper scripts, including sweep.py, rank-ads.py, and make-picker.py, to manage the advertisement data lifecycle. It also utilizes common system utilities like curl, jq, and find for networking and file processing.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to fetch media assets (MP4s and JPEGs) from external Meta CDN URLs and communicates with the vendor's API at api.novoads.ai.
[CREDENTIALS_UNSAFE]: Authentication is handled by reading a NOVOADS_API_KEY from a .env file at the repository root, following standard practices for local tool configuration.
[DATA_EXFILTRATION]: The skill transmits authentication credentials to the vendor's API endpoint (api.novoads.ai) to authorize the retrieval of competitive advertisement data.
[PROMPT_INJECTION]: The skill ingests untrusted content (ad copy and page names) from the Meta Ad Library. This external data is processed and presented to the agent, creating a surface for indirect prompt injection.
Ingestion points: Data retrieved from the POST /v1/competitor-ads endpoint and stored in sweep.json.
Boundary markers: The instructions do not define specific delimiters or warnings to isolate untrusted advertisement content within the agent's context.
Capability inventory: The skill has the capability to execute local scripts and perform network downloads based on its processing of the ingested data.
Sanitization: No explicit sanitization or validation of the advertisement text was identified in the provided workflow descriptions.

spy-competitor-ads