The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and summarize data from untrusted sources (files within a codebase, such as README.md, package manifests, and source code). This creates an attack surface where a malicious project could embed instructions to manipulate the agent's output or actions. * Ingestion points: Discovery steps in SKILL.md execute commands like cat, grep, and head on various codebase files to extract project metadata and architecture. * Capability inventory: The agent is granted access to Read, Bash, Grep, and Glob tools for codebase analysis. * Boundary markers: The instructions lack explicit delimiting markers or warnings to the agent to treat external codebase content as data rather than instructions. * Sanitization: The skill provides no mechanisms to filter or sanitize potential prompt injection patterns found within the project files before they are processed by the agent's reasoning engine.

explain-codebase