skills/kongshan001/opensource-project-learning-skill/opensource-project-learning/Gen Agent Trust Hub
opensource-project-learning
Pass
Audited by Gen Agent Trust Hub on May 10, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill guides the agent to perform local file system analysis using provided bash scripts (
analyze-project.sh,analyze-structure.sh) and suggests executing project-specific installation or run commands (e.g., in Stage 0 and Stage 3) based on the project's own documentation and requirements.\n- [EXTERNAL_DOWNLOADS]: As part of its learning workflow, the skill encourages the installation of project dependencies via standard package managers (npm, pip, cargo, etc.), which involves downloading external code from public registries.\n- [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it is designed to read and interpret untrusted external content (README files, source code, and configuration files) from third-party repositories to explain project functionality to the user.\n - Ingestion points: SKILL.md (Stage 1 and 2), scripts/analyze-project.sh.\n
- Boundary markers: Absent in instructions regarding external content processing, though the skill uses a step-by-step confirmation approach with the user.\n
- Capability inventory: Full file system access and shell command execution capabilities are leveraged to analyze and run the target projects.\n
- Sanitization: No explicit sanitization or filtering of external project content is performed before processing.
Audit Metadata