The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted social media data (such as comments, profiles, and post content) scraped via Apify Actors. This data is displayed in the chat and summarized by the agent, creating a surface for indirect prompt injection attacks where malicious instructions hidden in social media content could influence the agent's behavior.
Ingestion points: Data is fetched from Apify's dataset API in run_actor.js and presented to the agent for summarization in Step 5 of SKILL.md.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are used when interpolating external data into the agent's context.
Capability inventory: The agent has the ability to execute shell commands (mcpc, node) and write files to the local system.
Sanitization: While run_actor.js performs basic escaping to prevent CSV formula injection, no sanitization or filtering is applied to protect the LLM from instructions embedded within the scraped text.
[COMMAND_EXECUTION]: The instructions in SKILL.md direct the agent to execute shell commands involving the mcpc CLI tool and the provided run_actor.js script. These commands include environment variable extraction using grep and xargs and piping output to jq.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to Apify's official API domains (api.apify.com, mcp.apify.com) to fetch actor schemas and download scraping results. It also requires the installation of the @apify/mcpc global NPM package.
[DATA_EXPOSURE]: The skill accesses the APIFY_TOKEN credential stored in a .env file to authenticate requests to Apify services.

apify-audience-analysis