The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill contains instructions for the agent to read and extract sensitive API keys directly from the user's shell configuration file (~/.zshrc).
Evidence: SKILL.md and references/trend-research.md include bash commands like grep 'VIRLO_API_KEY' ~/.zshrc | grep -o '"[^"]*"' | tr -d '"' to programmatically harvest credentials.
[DATA_EXFILTRATION]: The skill performs network operations to external APIs (api.virlo.ai and api.elevenlabs.io) using credentials extracted from the local system.
Evidence: Multiple curl commands in references/trend-research.md and references/video-patterns.md send request data and tokens to these external endpoints.
[PROMPT_INJECTION]: The skill is vulnerable to Indirect Prompt Injection because it ingests untrusted data from external sources (Virlo API and WebSearch) and uses it to generate content.
Ingestion points: api.virlo.ai/trends/digest and generic WebSearch queries for trending topics in references/trend-research.md.
Boundary markers: None identified. There are no instructions for the agent to ignore or delimit potentially malicious instructions embedded in the retrieved trend data.
Capability inventory: The skill uses curl for network access, grep for file reading, and generates various file types (PSEO pages, video scripts, social posts).
Sanitization: No sanitization or validation of the ingested trend data is mentioned before it is interpolated into content generation prompts.

trend-to-content