The Agent Skills Directory

[DATA_EXFILTRATION]: The skill includes hardcoded absolute file paths (C:/Users/gurusharan.gupta/.claude and C:/Users/gurusharan.gupta/Agents/Claude Code) which expose a specific local username and system structure profile.\n- [COMMAND_EXECUTION]: The skill uses Bash to perform broad inspections of the local ~/.claude/ configuration directory and its subdirectories, including reading the settings.json file. While consistent with the audit purpose, this involves processing security-sensitive configuration files.\n- [PROMPT_INJECTION]: The skill processes content from various local files, creating a surface for indirect prompt injection where malicious content in those files could influence the audit.\n
Ingestion points: Reads manifest.json, CLAUDE.md, and all SKILL.md files located in the ~/.claude/skills/ directory (implemented in SKILL.md).\n
Boundary markers: Absent. The skill does not use delimiters to wrap the content of audited files or instructions to ignore embedded commands.\n
Capability inventory: The skill uses Bash, Read, Glob, and Grep across all audit categories in SKILL.md.\n
Sanitization: Absent. Files are processed as raw text without validation or filtering.

harness-audit