pr-review-workflow

Warn

Audited by Gen Agent Trust Hub on Mar 2, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The core rules in AGENTS.md explicitly permit the agent to use sudo directly if elevated privileges are required for a task, which increases the potential impact of command-line operations.
  • [EXTERNAL_DOWNLOADS]: Both the README.md and the rule composition in AGENTS.md recommend the global installation of various CLI tools from the author's organization via npm, specifically compose-agentsmd, @metyatech/task-tracker, and @metyatech/thread-inbox.
  • [REMOTE_CODE_EXECUTION]: The skill utilizes a dynamic rule-loading mechanism where compose-agentsmd fetches and regenerates the agent's behavioral instructions (AGENTS.md) from a remote GitHub repository (github:metyatech/agent-rules@HEAD). This allows for the runtime modification of the agent's logic from an external source.
  • [PROMPT_INJECTION]: The skill processes untrusted external data in the form of PR review feedback, creating an indirect prompt injection surface.
  • Ingestion points: PR review threads, comments, and GitHub notifications as specified in SKILL.md.
  • Boundary markers: Absent; the skill does not provide instructions to use delimiters or to treat the feedback as untrusted data.
  • Capability inventory: Mutative GitHub API calls (adding/removing reviewers, deleting notifications), git operations (merging, branch deletion), and sudo command execution.
  • Sanitization: Absent; there is no mention of validation, escaping, or filtering of the content retrieved from PR comments.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 2, 2026, 01:29 AM