The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from news stories and competitor ad libraries to generate ad concepts. 1. Ingestion points: News outlets, Telegram channels, and public ad libraries (Meta, Google, LinkedIn) as defined in SKILL.md and references/ad-libraries.md. 2. Boundary markers: The skill employs 'boundary of truth' constraints and specific rules to verify stories with independent sources. 3. Capability inventory: Generates ad copy, video spines, and executes shell commands for video processing. 4. Sanitization: Implements a publication gate involving independent model reviews and a 'cold-viewer' check to identify defects or misleading content.
[COMMAND_EXECUTION]: The skill provides instructions for the agent to execute shell commands for media manipulation. Evidence: references/synthetic-panel.md includes a command to extract frames from video files using ffmpeg for analysis.
[EXTERNAL_DOWNLOADS]: The skill facilitates data retrieval from well-known and trusted external market transparency services. Evidence: It accesses official platforms such as the Meta Ad Library, Google Ads Transparency Center, and LinkedIn Ad Library to scan competitor advertisements.

newsjack-ads