The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses user-supplied arguments to construct shell commands for Git and the GitHub CLI.
Evidence: In SKILL.md, the arguments provided via $ARGUMENTS are interpolated into shell execution patterns such as gh pr diff <PR_NUMBER> --name-only and git diff <default_branch>...HEAD. If the agent does not strictly validate or sanitize these arguments before execution, an attacker could use shell metacharacters (e.g., ;, &&, |) to execute arbitrary commands on the host system.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it reads untrusted data from external sources and incorporates it into a prompt for another agent.
Ingestion points: The skill reads file diffs via git diff and PR metadata (titles and descriptions) via gh pr view and gh pr diff.
Boundary markers: There are no explicit boundary markers or instructions to the downstream agent to ignore instructions embedded within the reviewed code or PR metadata.
Capability inventory: The skill uses Bash and the Task tool (invoking majestic-engineer:workflow:quality-gate).
Sanitization: The skill does not implement sanitization or filtering for the content fetched from the repository or GitHub API before passing it to the secondary agent. A malicious user could craft a Pull Request title or code comment designed to hijack the review agent's logic.

quality-gate