report-upstream-bug
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill performs shell operations to manage code reproductions and install project dependencies.
- Evidence: Git worktree and fetch commands are executed against local and remote paths in SKILL.md (Step 2).
- Evidence: The pnpm install command is used to install external dependencies within the reproduction environment.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to search for and read external GitHub issues and pull requests, which may contain malicious instructions designed to manipulate agent behavior.
- Ingestion points: External GitHub issue and PR content retrieved in Step 1 of SKILL.md.
- Boundary markers: No clear delimiters are specified to separate external issue content from agent instructions.
- Capability inventory: The agent can execute shell commands (git, pnpm), write files (README), and create tickets via tool calls (/create-ticket).
- Sanitization: No explicit sanitization or validation of the ingested external content is mentioned.
- [EXTERNAL_DOWNLOADS]: The skill interacts with external repositories and package registries.
- Evidence: Fetches and clones code from github.com/jd-solanki/reproductions (Vendor Resource).
- Evidence: Downloads Node.js packages via pnpm install.
Audit Metadata