The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill performs read-only HTTPS requests to the well-known service ipinfo.io to determine the terminal's exit IP and geographic metadata. This behavior is consistent with the skill's auditing purpose and involves no untrusted code execution.
[COMMAND_EXECUTION]: The skill executes local system utilities (such as ps, scutil, npm, git, and curl) via Python's subprocess module to inspect running processes and environment configurations. The arguments are controlled and specific to gathering audit data for Claude applications.
[INDIRECT_PROMPT_INJECTION]: The skill exposes an ingestion surface by reading and reporting on content from external network auditing websites (e.g., IPLeak, BrowserLeaks). 1. Ingestion points: Data is read from external audit pages as described in references/browser-and-report.md. 2. Boundary markers: The skill uses absolute path variables (e.g., SKILL_DIR). 3. Capability inventory: The skill performs file moves/writes, local command execution, and local browser operations. 4. Sanitization: The claude_check_report.py script employs html.escape and IP masking, while claude_network_check.py includes logic to redact detected secrets and credentials from shell profiles and environment variables.

claude-local-reset