The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill exposes a customMapFunction parameter described as a "JavaScript function to transform each output object." This allows for the execution of arbitrary code logic within the data processing pipeline.
[COMMAND_EXECUTION]: The skill provides instructions to run shell commands using node scripts/run_actor.js and curl. These commands involve sensitive operations like writing files to the local system (--output YYYY-MM-DD_results.csv) and making network requests.
[DATA_EXFILTRATION]: The instructions explicitly mention using the APIFY_TOKEN environment variable and provide curl commands that send this sensitive credential to api.apify.com. While this is intended for the service's functionality, it represents a pattern of transmitting credentials over the network.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external, untrusted sources (Instagram and TikTok via scrapers). There is a documented attack surface where malicious content embedded in social media posts could influence the agent's behavior during the classification and scoring steps (Step 3 and 4).
Ingestion points: Data enters the context from the Apify dataset items fetched via curl or run_actor.js.
Boundary markers: None identified in the instructions for the agent to distinguish between platform metadata and potential malicious content in social media bios or posts.
Capability inventory: The skill uses node execution, curl for network requests, and file-writing capabilities.
Sanitization: There is no mention of sanitization or filtering of the scraped content before processing.

finding-affiliate-marketers-on-social-media