Clawpilot Skill Installation Advisor

What Problem It Solves

When a user knows the outcome but not which skill to install, Clawpilot turns a natural-language task into a skill selection recommendation. It detects intent, narrows candidates to 1-3 skills, compares tradeoffs, labels risk, and proposes an installation order. This is useful when the task is vague, skill names are unknown, or the user needs to avoid choosing a high-risk capability without context.

How It Works

The core input is a prompt such as “I want to track where my package is.” Clawpilot extracts a task category from its bundled intent map, such as logistics, mental health, legal support, or shopping. It then returns candidate skills, for example logistics, burnout-checkin, or clause-redraft, and explains the main difference between them.

Key steps:
- Intent detection: maps phrases like “draft a rental contract” or “find a legal skill” to a task category
- Candidate recommendation: returns a short list instead of asserting a single best answer
- Risk labeling: marks options as low, medium, high, or pending
- Installation guidance: recommends a basic capability first and a narrower add-on later only when needed
- Mandatory disclaimers: medium and high-risk outputs must note that the skill does not replace legal, medical, or professional advice

Boundaries and Caveats

Clawpilot v1 is an advisor, not an installer, and it does not automatically install skills. It explicitly avoids medical diagnosis, investment advice, international logistics, and technical implementation consulting. Its shipped intent mapping is currently strongest for Chinese-language queries, so English prompts should still be framed as skill discovery, workflow selection, or risk checking. If the request is outside supported scope, it should decline and ask the user to rephrase toward a concrete skill-discovery task.