fetching-github-issue

Warn

Audited by Gen Agent Trust Hub on Apr 17, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using the GitHub CLI (gh).\n
  • Evidence: In subagents/issue-retriever.md, the agent is directed to construct and run commands like gh issue view <ISSUE_NUMBER> --repo <owner>/<repo> and gh api calls using identifiers derived from user input.\n
  • Risk: There is a potential for command injection if the user-provided URL or identifiers contain shell metacharacters and the agent does not correctly sanitize or escape them before execution.\n
  • Mitigation: The skill includes explicit instructions for the agent to validate input coordinates and normalize identifiers before performing any GitHub operations.\n- [PROMPT_INJECTION]: The skill processes untrusted data from an external source (GitHub), serving as a potential vector for indirect prompt injection targeting subsequent workflow steps.\n
  • Ingestion points: The skill retrieves issue bodies and comments from the GitHub API using gh (documented in subagents/issue-retriever.md).\n
  • Boundary markers: Output is structured according to a Markdown snapshot template (subagents/issue-retriever-template.md).\n
  • Capability inventory: The agent has the ability to execute shell commands and write to the local filesystem (docs/).\n
  • Sanitization: The skill mitigates structural confusion by instructing the agent to rewrite source Markdown headings (e.g., ##) as bold labels to prevent them from colliding with the snapshot's reserved sections.\n
  • Risk: Malicious instructions embedded in GitHub issues or comments could influence the behavior of other agents that consume the generated artifact.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 17, 2026, 09:12 AM