automate-github-issues
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The
scripts/setup.shscript andSKILL.mdinstructions include a command to install the Bun runtime viacurl -fsSL https://bun.sh/install | bash. This downloads and executes a script from the official Bun domain, which is a well-known service. - [COMMAND_EXECUTION]: The
scripts/github/git.tsutility uses theexecfunction to rungitcommands locally, such asgit remote get-urlandgit rev-parse, to retrieve repository information needed for GitHub API interactions. - [INDIRECT_PROMPT_INJECTION]: The skill implements a multi-agent pipeline where external data (GitHub issues) is used to generate prompts for subsequent AI agents.
- Ingestion points:
scripts/fleet-analyze.tsfetches open issues from GitHub repositories. - Boundary markers: The prompt template in
scripts/prompts/analyze-issues.tsdefines explicit "FILE BOUNDARY" rules to constrain agent actions. - Capability inventory: Dispatched agents utilize the Jules SDK to write code and manage PRs across the repository.
- Sanitization: Untrusted issue data is formatted into markdown templates to provide structure.
Audit Metadata