The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes curl to interface with Apify's API for scraping operations and status polling. It also leverages local system utilities including ffmpeg, ffprobe, and whisper-cli to handle video file analysis and transcription tasks.
[EXTERNAL_DOWNLOADS]: Video advertisements are downloaded from Meta's content delivery networks to the local /tmp/ad_videos/ directory. These files are processed locally and then deleted once transcription and metadata extraction are complete.
[DATA_EXFILTRATION]: Advertisement metadata, creative URLs, and transcripts are transmitted to the user's Airtable base. This network activity is central to the skill's purpose of maintaining a competitor database and targets well-known service providers.
[PROMPT_INJECTION]: The classification logic in Step 6 processes untrusted content (ad copy and transcripts) from external sources, which represents an indirect prompt injection surface.
Ingestion points: Competitor advertisement text, titles, and generated transcripts from the Meta Ad Library (SKILL.md).
Boundary markers: No specific delimiters or instructions to ignore embedded commands are used when presenting external data for classification.
Capability inventory: The skill possesses capabilities for shell command execution (curl, ffmpeg) and network communication with Airtable and Apify.
Sanitization: The skill does not implement explicit sanitization or filtering of the advertisement data before passing it to the model for classification.
[CREDENTIALS_UNSAFE]: The skill instructs the agent to retrieve API tokens from the project's .mcp.json or a .env file. This approach avoids hardcoding sensitive credentials and adheres to standard security practices for secret management.

spy-on-competitor-meta-ads