The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes external source code, which introduces a surface for indirect prompt injection if malicious instructions are embedded in comments or strings within the target codebase.
Ingestion points: The skill reads source files, project configurations (e.g., csproj, package.json), and symbols using Read, Grep, Glob, and Serena MCP tools.
Boundary markers: The skill does not use specific delimiters for data ingestion; however, it includes strong instructional principles such as "Prioritize actual code over comments" and "Behavior over Intent," which serve as effective mitigations against following instructions hidden in code comments.
Capability inventory: The skill is permitted to use Write for creating reports and a manifest.json, and Bash for creating the docs/ directory structure. It cannot execute the code it analyzes.
Sanitization: The skill does not perform sanitization as its purpose is the verbatim documentation of implemented code structures and behaviors.
[SAFE]: The skill follows security best practices by explicitly prohibiting high-risk actions such as code execution, build commands, and modifications to the source codebase. It uses relative paths and localizes all output to a dedicated documentation directory.

re-structure-analysis