The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and cross-reference text from external, untrusted sources such as research papers. This introduces a surface where malicious instructions embedded in a source document could attempt to influence the agent's behavior.
Ingestion points: Step 1 in SKILL.md explicitly directs the agent to collect and extract exact limitation statements from 5-10 external papers.
Boundary markers: The protocol lacks instructions to use delimiters or ignore-embedded-instruction markers when handling the extracted text.
Capability inventory: While the skill itself does not define scripts, the lack of tool restrictions in the frontmatter means the agent operates with its full default capabilities, which may include file system or network access.
Sanitization: The instructions do not provide for any sanitization or validation of the content extracted from external sources.

paradox-sifter