The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it is designed to ingest and process untrusted data from external repositories, pull request descriptions, and source code files.
Ingestion points: Processes repository source code, configuration files, test scripts, and GitHub PR diffs/comments (SKILL.md, 'Always audit against the original source' section).
Boundary markers: The instructions do not define specific delimiters or instructions to treat the analyzed content as untrusted data, which could allow malicious instructions embedded in code comments or PR metadata to influence the agent's output.
Capability inventory: The skill utilizes file system access, git operations, and CLI interactions to retrieve the audit material.
Sanitization: There is no evidence of automated sanitization or filtering of the content being audited before it is processed by the LLM.
[COMMAND_EXECUTION]: The skill instructions direct the agent to perform standard development operations using system CLI tools.
Evidence: Instructions include using git fetch, git checkout, gh pr checkout, and git clone to acquire source code for auditing (SKILL.md).
Context: These commands are used as intended for the primary purpose of the skill (auditing code) and utilize established, well-known developer tools.
[SAFE]: The skill implements operational best practices by requiring grounded analysis of actual source code and prioritizing flows based on expected user impact. It identifies specific vendor-related resources (Altinity repositories) for its context.

usability-audit