sentry-code-review

Pass

Audited by Gen Agent Trust Hub on Mar 20, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. It is explicitly instructed to parse a section titled 'Prompt for AI Agent' within GitHub Pull Request comments and follow the instructions therein to implement code fixes. If an attacker or a compromised bot provides a crafted comment, the agent could be manipulated into introducing vulnerabilities, backdoors, or performing unauthorized code modifications.
  • Ingestion points: The skill ingests untrusted data from GitHub PR comments via the 'gh api repos/{owner}/{repo}/pulls/<PR_NUMBER>/comments' command.
  • Boundary markers: While the skill expects a specific Markdown/HTML structure (e.g., details tags), it lacks explicit instructions to ignore or sanitize adversarial commands embedded within the 'Prompt for AI Agent' section.
  • Capability inventory: Across its workflow, the skill relies on 'Read' and 'Edit' tools to view and modify the local repository's source code based on the parsed instructions.
  • Sanitization: There is no mechanism for sanitizing or validating the content retrieved from the GitHub API before the agent adopts it as functional instructions.
  • [COMMAND_EXECUTION]: The skill executes the 'gh' (GitHub CLI) tool to retrieve pull request metadata and comments. This is a legitimate operation used to fulfill the skill's primary purpose of code review automation.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 20, 2026, 04:57 PM