The Agent Skills Directory

[METADATA_POISONING]: Discrepancy in author information. The skill metadata in SKILL.md identifies the author as 'K-Dense Inc.', which conflicts with the provided author context 'LeonChaoX'.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests untrusted data from external research APIs (Parallel and Perplexity) which are then processed by the agent. * Ingestion points: Research results retrieved from external API endpoints in research_lookup.py and lookup.py. * Boundary markers: Uses markdown formatting and headers to separate content. * Capability inventory: Requests Bash, Write, and Edit tools in SKILL.md. * Sanitization: Does not perform explicit sanitization of research content before presenting it to the agent.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch research data from external services. * Communicates with api.parallel.ai and openrouter.ai (well-known service).
[COMMAND_EXECUTION]: The skill requests broad environment permissions (Bash, Write, Edit) in SKILL.md. This allows for arbitrary command execution and file system modification by the agent based on instructions received from processed data.

research-lookup