The Agent Skills Directory

[PROMPT_INJECTION]: The skill's workflow for researching and reporting on external technologies introduces an indirect prompt injection surface. * Ingestion points: Reads local build configuration files (e.g., package.json, pyproject.toml, Cargo.toml) and untrusted data from web search results (SKILL.md). * Boundary markers: The instructions do not specify the use of delimiters or clear directives to ignore instructions contained within the ingested project files or search data (SKILL.md). * Capability inventory: The skill allows the agent to write files to the local 'reports/' directory based on gathered information (SKILL.md). * Sanitization: There is no instruction to sanitize or escape the external content before it is included in the final report (SKILL.md).

tech-research