The Agent Skills Directory

[PROMPT_INJECTION]: The skill incorporates specific security guidelines to handle indirect prompt injection. It instructs the agent to treat all scraped data, such as Instagram captions and Google Drive files, as material for analysis rather than instructions to be followed.
Ingestion points: Data is retrieved from Instagram via Apify and ScrapeCreators, Gmail, Slack DMs, and Google Drive records.
Boundary markers: The skill uses a clear directive at the beginning to define scraped data as untrusted and provides a behavioral constraint for any text that addresses the agent directly.
Capability inventory: The skill interacts with Airtable to update deal records and uses external APIs for web scraping and content transcription.
Sanitization: The instructions direct the agent to quote any text addressed to it and cease execution, preventing accidental obedience to embedded prompts.
[DATA_EXFILTRATION]: The skill accesses sensitive communication and storage platforms including Gmail, Slack, and Google Drive to retrieve deal terms and briefs. These operations are essential to the skill's primary function of brand research and utilize well-known services. The skill adheres to security best practices by recommending the use of environment variables for API key management and specifying read-only access for messaging platforms.
[EXTERNAL_DOWNLOADS]: The skill utilizes external scraping services including Apify and ScrapeCreators to gather engagement metrics and transcripts from Instagram. These are well-known services used for their intended purpose in the context of market research.

brand-research