The Agent Skills Directory

[SAFE]: The skill is a professional-grade functional utility for web archiving and offline reading.
[COMMAND_EXECUTION]: Uses subprocess.run in scripts/invoke.py and scripts/mirror.py to execute legitimate system tools (uv, agent-browser) required for its operation. The logic is used to manage the execution environment and perform browser-based rendering, which is consistent with the skill's stated purpose.
[DATA_EXPOSURE_EXFILTRATION]: The skill performs network requests to fetch web content as explicitly requested by the user. Analysis found no evidence of hardcoded credentials, harvesting of local sensitive files (e.g., SSH keys, AWS configs), or exfiltration of user data to unauthorized third parties.
[INDIRECT_PROMPT_INJECTION]: While the skill ingests untrusted data from the web, it includes robust sanitization measures. Specifically, scripts/mirror.py (via the without_scripts and rewrite_html functions) strips all <script> tags, script-supporting network hints, and integrity attributes from the mirrored content. Additionally, the skill configuration includes disable-model-invocation: true, which reduces the attack surface for multi-step injection chains.
[DYNAMIC_EXECUTION]: The skill uses uv run to execute its own scripts and the platform's manager engine. The paths for the manager are searched within standard agent configuration directories (e.g., ~/.claude/skills, ~/.cursor/skills), which is part of the vendor's intended architecture for skill discovery.

mirror