The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and process content from potentially untrusted external sources, including PDFs, Word files, web pages, URLs, and APIs. This creates a surface for indirect prompt injection attacks where malicious instructions hidden in these sources could influence the agent's behavior during the data creation process.
Ingestion points: intake of documents, web content, and API data as described in SKILL.md.
Boundary markers: No specific delimiters or safety instructions are provided to the agent to treat external content as data rather than instructions.
Capability inventory: The skill utilizes npx skills for runtime dependency resolution and executes local CLI commands via the tiangong-lca tool suite.
Sanitization: The workflow lacks explicit steps for sanitizing or validating the text content retrieved from external sources before processing it into the evidence dossier.
[EXTERNAL_DOWNLOADS]: The skill resolves runtime requirements using npx skills use, fetching tools from the tiangong-ai GitHub repository. This behavior is documented as part of the standard evidence-gathering workflow for scientific literature.
[REMOTE_CODE_EXECUTION]: The skill uses npx skills use https://github.com/tiangong-ai/skills to dynamically load and execute specialized evidence-retrieval tools at runtime. This allows the agent to extend its capabilities based on the specific evidence channel required.

source-evidence-dataset-development