The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to ingest and process large amounts of untrusted data from the internet, creating a significant surface for indirect prompt injection. It aggregates content from Reddit, X, YouTube transcripts, Hacker News, and Polymarket, then provides this content to the agent for synthesis and prompt generation.
Ingestion points: Research results compiled by scripts/last30days.py from multiple third-party sources.
Boundary markers: Output is organized with markdown headers (e.g., ### Reddit Threads), but no strict delimiters or explicit instructions to ignore nested commands are provided.
Capability inventory: The skill has access to Bash, Write, and WebSearch tools, enabling it to execute local code and modify files based on potentially malicious ingested content.
Sanitization: There is no evident sanitization or filtering of the ingested content to prevent instructions embedded in web data from being interpreted by the agent.
[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to execute its core Python orchestrator and Node.js search client. While these scripts are part of the skill's own package, this execution model is central to its operation.
[EXTERNAL_DOWNLOADS]: The skill relies on several well-known external services for data retrieval. It also recommends the installation of yt-dlp for YouTube research. These references are to trusted organizations and reputable technology providers.

last30days