The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from multiple untrusted sources including Reddit, X/Twitter, Hacker News, and general web articles. This creates a surface for indirect prompt injection where malicious instructions embedded in the source data could influence the agent's behavior.
Ingestion points: SKILL.md fetches data via curl and WebFetch from various social media and web platforms in steps 2, 3, 4, and 6.
Boundary markers: The skill does not use explicit delimiters when passing external content to the models, though it implements a multi-step summarization process that reduces the risk of direct instruction following.
Capability inventory: The skill can execute shell commands (curl), write to the filesystem (output/, memory/), and send notifications via ./notify.
Sanitization: The skill includes an explicit instruction to discard any source containing common prompt injection patterns like "ignore previous instructions" or "you are now..." in step 6.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to several external platforms for research data.
Sources: The skill fetches data from well-known services including Reddit (old.reddit.com), X/Twitter (api.x.ai), Hacker News (hn.algolia.com), and prediction markets (gamma-api.polymarket.com, api.elections.kalshi.com). These requests are central to the skill's primary research purpose.
[COMMAND_EXECUTION]: The skill uses shell commands for environment setup, parameter parsing, and data fetching. It utilizes standard Linux utilities such as curl, jq, sed, tr, and date. It employs a specialized script ./secretcurl to handle the XAI_API_KEY for requests to api.x.ai, which is a standard method for managing API credentials.

last30