The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from git diffs, GitHub issues, and Linear tickets. This data is interpolated into the prompts for sub-agents, such as the requirements-verifier, without the use of explicit boundary markers or sanitization, which could allow an attacker to manipulate the review outcome through crafted instructions in the source code or external task descriptions.
[EXTERNAL_DOWNLOADS]: The skill fetches requirements and issue details from well-known services like GitHub and Linear. This integration is a primary feature of the skill, used to cross-reference delivery against specifications.
[COMMAND_EXECUTION]: The skill executes local development commands including git, mkdir, and Elixir's mix tool. The verification-runner agent specifically executes mix test and mix compile, which involves running code from the repository being reviewed.
[COMMAND_EXECUTION]: Certain shell operations, such as command-line expansion of file lists from git diff for mix format, may be vulnerable to command injection if the repository contains maliciously crafted filenames with shell metacharacters.

review