Intent-based tool selection. Activate when you want to know which tools are relevant for the current user message (weather→exec, document→feishu_doc, search→...

This skill appears to do what it says: keyword-based recommendations for which tools to expose. Before installing or applying its suggestions: 1) Be cautious about following the README's 'gateway-level filtering' advice — it proposes patching OpenClaw distribution files (auth-profiles-*.js). Back up those files and review any code you paste into core bundles. 2) Note that the implementation always includes the 'exec' tool in recommendations; ensure your exec tool is safely sandboxed or disabled if you don't want agents running shell commands. 3) Verify and inspect any external links or code snippets (the README references a GitHub path that looks like a placeholder). 4) If you plan to apply the gateway patch, restrict who can perform that change and test in a non-production environment. If you want a lower-risk approach, use the skill's get_recommended_tools output in your agent logic without modifying core OpenClaw files.

Type: OpenClaw Skill Name: tooldyn Version: 1.0.0 The dynamic-tool-policy skill is a utility designed to optimize tool selection for AI agents by mapping user intent to specific tools (e.g., weather to 'exec', search to 'web_search'). The implementation in index.js is a straightforward keyword-based router with no requested permissions and no external network or file system access. While it encourages the use of the high-privilege 'exec' tool for weather queries (via wttr.in) and mentions manual patching of OpenClaw core files for advanced optimization in the README.md, these are presented as legitimate architectural workarounds for token optimization in small-model environments and do not exhibit malicious intent.