The Agent Skills Directory

[SAFE]: The skill consists entirely of markdown documentation and instructional guidelines. It does not ship with any scripts (.py, .js, .sh), binaries, or configuration files that execute commands.
[SAFE]: No obfuscation, hidden characters, or multi-layer encoding were found in the provided files. All instructions are clear and transparent.
[SAFE]: The skill does not perform any direct network operations or external downloads. While it references external skills like literature-search, it does not provide untrusted URLs or unverified dependencies.
[INDIRECT_PROMPT_INJECTION]: As a retrieval-augmented generation (RAG) framework, the skill inherently possesses an indirect prompt injection surface.
Ingestion points: Untrusted data enters the agent context via the user query and the results of the literature-search and scienceclaw-ie tools mentioned in SKILL.md.
Boundary markers: The skill lacks explicit instructions for the agent to ignore potentially malicious instructions embedded within retrieved scientific papers, though it includes a 'Zero-Hallucination Rule'.
Capability inventory: The skill itself has no direct capabilities, but it instructs the agent to delegate tasks to the code-execution and literature-search tools.
Sanitization: No explicit sanitization or validation of the retrieved evidence is described in the synthesis methodology.

scienceclaw-qa