The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill acts as an analyst for long-form content, which is a known surface for indirect prompt injection. Maliciously crafted articles or web pages could contain hidden instructions designed to hijack the agent's logic or bias its synthesis. (1) Ingestion points: The skill processes user-provided long-form content and URLs as specified in the initialization step of SKILL.md, as well as external content retrieved via the web_search tool during Level 4 (Research) analysis. (2) Boundary markers: Absent; the instructions do not mandate the use of delimiters (e.g., XML tags or triple backticks) or specific 'ignore' directives to prevent the agent from following instructions embedded in the external content. (3) Capability inventory: The skill utilizes the web_search tool for cross-source comparison. No high-risk capabilities such as file system modification, shell command execution, or credential access were identified. (4) Sanitization: No explicit sanitization or validation of the ingested external content is described in the analysis workflow.

deep-reading-analyst