The Agent Skills Directory

[Data Exposure] (SAFE): The tool reads source code from the target directory for analysis. While it does not explicitly exclude secret files like .env or .ssh keys by default, its behavior is consistent with its primary purpose of codebase auditing. Standard exclusions for .git, node_modules, and build directories are implemented.
[Indirect Prompt Injection] (SAFE): The skill processes untrusted source code and extracts snippets into reports, presenting a theoretical surface for indirect injection if a downstream agent follows instructions embedded in analyzed comments.
Ingestion points: Source files scanned via glob patterns and read into memory in CodeAnalyzer.js and SimpleDeduper.js.
Boundary markers: Absent; code snippets are interpolated directly into JSON, HTML, or console outputs.
Capability inventory: File reading (fs.readFile), file writing (fs.writeFile), and file deletion (fs.unlink for cache management) are used for legitimate functionality.
Sanitization: None; the tool treats code content as text for duplicate and structure analysis.
[External Downloads] (SAFE): Dependencies in package.json are standard, well-known libraries from the npm registry (e.g., commander, glob, tree-sitter, chalk). No suspicious or unversioned remote scripts are executed at runtime.
[Command Execution] (SAFE): The skill performs static analysis and does not use eval, exec, or spawn to run arbitrary commands or user-provided scripts.

code-dedup