repository-debt-analysis
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted repository data, including filenames, git history, and source code content, which provides a surface for indirect prompt injection.
- Ingestion points: Untrusted data enters the context via repository file listings (Step 1), git log extraction (Step 2), and file content analysis (Step 3).
- Capability inventory: The skill utilizes shell commands (git, wc, sort, uniq, xargs, grep) to process this data. It includes capabilities for broad file reading and git history inspection.
- Boundary markers: The skill explicitly instructs the agent to establish an analysis scope and exclude specific categories like submodules and generated code (Step 1), though it does not mandate specific delimiters for content.
- Sanitization: The instructions do not specify sanitization or escaping techniques for filenames or git metadata before they are passed to shell utilities.
- [COMMAND_EXECUTION]: The skill uses local shell commands to gather statistics and evidence from the repository being analyzed.
- Evidence: Several bash command sequences are provided in Steps 1 and 2 to count files, check submodules, analyze change frequency, and identify large files. The skill is explicitly defined as read-only and prohibits modifying, fixing, or creating new files, which limits the risk associated with command execution.
Audit Metadata