hunt-source-leak
Fail
Audited by Gen Agent Trust Hub on Nov 12, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill makes extensive use of shell commands (including curl, grep, and git) to perform reconnaissance tasks against target infrastructure.
- [EXTERNAL_DOWNLOADS]: The skill installs several third-party security tools and libraries at runtime, including git-dumper and ds_store via pip3, and source-map-explorer and unwebpack-sourcemap via npm.
- [DYNAMIC_EXECUTION]: The skill uses inline Python scripts (python3 -c) to process and manipulate data retrieved from remote targets. This includes parsing JSON, creating directories, and writing files to the local system under /tmp/sourcemap_extract/.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from remote web servers defined by the $TARGET variable. While the processing is handled by scripts, the resulting information (such as extracted secrets and internal endpoints) is brought into the agent's context, creating a surface where malicious content from a target could influence the agent.
- Ingestion points: Data retrieved via curl from various paths on the target (e.g., .env, .map files, swagger.json).
- Boundary markers: Absent; data is processed and displayed directly to the agent.
- Capability inventory: Shell command execution, network requests, file writing, and package installation.
- Sanitization: Basic path traversal prevention is applied when extracting source maps, but general input validation for remote data is minimal.
Recommendations
- HIGH: Downloads and executes remote code from: https://$TARGET/asset-manifest.json - DO NOT USE without thorough review
Audit Metadata