issue-triage
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from GitHub issues and comments, creating a surface for indirect prompt injection.
- Ingestion points: GitHub issue bodies and comments are fetched using
gh issue listandgh issue viewinSKILL.md(Phase 1 and Phase 2). - Boundary markers: The instructions lack specific boundary markers or instructions to the LLM to ignore potentially malicious content within the issue data.
- Capability inventory: The skill has significant capabilities, including executing shell commands (
gh issue comment/edit/close) and accessing the local clipboard via system tools (pbcopy,xclip, etc.). - Sanitization: There is no logic provided to sanitize or escape the untrusted GitHub data before it is processed by subagents or used in shell commands.
- [COMMAND_EXECUTION]: The shell implementation for executing triage actions is vulnerable to command injection.
- Evidence: In
Phase 3: Execution, the skill uses a shell heredoc (<<'TRIAGE_EOF') to pass comment text to theghCLI. An attacker could craft a GitHub issue comment that includes the delimiterTRIAGE_EOFfollowed by malicious shell commands, which would then be executed on the user's machine during the automated action phase. - Evidence: The clipboard logic in
Phase 1usesecho "{full triage tables}" | xclip, which is susceptible to shell injection if the table content contains characters that break out of the double-quoted string.
Audit Metadata