The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes tweets and conversations from Twitter, which is an untrusted external data source. This creates a surface for indirect prompt injection if malicious instructions embedded in tweets are processed by the agent without proper sanitization.
Ingestion points: Data enters through the apidojo/tweet-scraper output which is then classified and scored in SKILL.md.
Boundary markers: No explicit instruction delimiters or boundary markers are defined to separate external data from agent instructions.
Capability inventory: The skill uses node scripts/run_actor.js and curl for network operations, and supports data transformation via customMapFunction.
Sanitization: No evidence of sanitization or filtering for injection patterns in the ingested tweet content.
[DYNAMIC_EXECUTION]: The skill defines an input parameter customMapFunction described as a 'JavaScript function to transform each output object'. If this function is executed by the agent or the supporting scripts on the local environment, it represents a dynamic code execution risk.
[DATA_EXFILTRATION]: The skill requires an APIFY_TOKEN and performs network operations via curl and the run_actor.js script to communicate with api.apify.com. While these are directed at a well-known service (Apify), the pattern of fetching external data and processing it locally is documented. The instructions also suggest storing secrets in environment variables or .env files, which is a standard security practice.

monitoring-trending-topics-twitter-by-niche