spec-miner
Warn
Audited by Gen Agent Trust Hub on Apr 13, 2026
Risk Level: MEDIUMDATA_EXFILTRATIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill instructions and reference materials (references/analysis-checklist.md, references/analysis-process.md) explicitly direct the agent to search for and analyze sensitive files and patterns, including environment configuration files (.env), authentication guards, JWT implementations, and credential management services.
- [COMMAND_EXECUTION]: The skill requests and utilizes the Bash tool to perform structural analysis and pattern matching. While the documented examples focus on codebase exploration using Glob and Grep, the presence of shell execution capabilities on untrusted local environments represents a significant capability tier.
- [PROMPT_INJECTION]: The skill is inherently susceptible to indirect prompt injection due to its core workflow of reading and interpreting external, untrusted source code and documentation.
- Ingestion points: Codebase files, configuration modules, and documentation processed during the discovery and tracing phases (SKILL.md, references/analysis-process.md).
- Boundary markers: The prompt instructions lack specific delimiters or isolation techniques to separate ingested codebase content from instructions, increasing the risk of the agent obeying commands embedded in code comments or metadata.
- Capability inventory: The agent is granted access to Read, Grep, Glob, and Bash tools, providing multiple vectors for tool abuse if a malicious codebase is analyzed (SKILL.md).
- Sanitization: No explicit sanitization, filtering, or validation of ingested content is defined in the analysis process.
Audit Metadata