Skill Finder (Find ClawHub skills + Search Skills.sh)

• You want your agent to proactively suggest installable skills whenever a user describes missing functionality or asks how to automate a task.
• You need to compare options across both the ClawHub curated registry and the broader Skills.sh open ecosystem before committing to an install.
• You want a consistent, preference-aware discovery workflow — the skill remembers liked/passed skills and your preferred search source across sessions.
• A user asks "what should I install for X?" and you want a structured, evaluated answer rather than a raw search dump.

• You only ever need to search a single, already-known registry and have no interest in cross-ecosystem comparison.
• Your environment cannot run npx commands (required for both clawhub and skills CLIs).

Searches both the ClawHub curated registry (npx clawhub search) and the open Skills.sh ecosystem (npx skills find) by default. Results from both sources are compared side-by-side so you always see the strongest cross-ecosystem option, not just the first match found.

Activates on signals like "how do I do X?" or "can this be automated?" — not only on explicit "find a skill" requests. This closes the gap between a user describing a need and discovering that an installable skill already solves it.

Every candidate is assessed on description clarity, download count, last-update recency, and author reputation before being recommended. Results are ranked with a clear winner and explicit tradeoff notes — never a raw list dump.

Stores your preferred source mode (both, clawhub, or skills.sh), liked skills, passed skills, and stated preferences in ~/skill-finder/memory.md. Future searches respect these settings automatically.

Never auto-installs. Every recommendation requires explicit user consent, skips force-install flags, avoids silent global scope changes, and surfaces scanner warnings before proceeding.

Translates vague user requests into precise search terms — e.g., "help with PDFs" becomes separate targeted queries for editing, generating, or parsing — reducing noise and improving result relevance.

A user says "I need to extract text from PDFs." Skill Finder detects the capability gap, searches both ecosystems for pdf parse and pdf extract, evaluates the top candidates, and presents a ranked recommendation with the exact install command — without the user ever typing "find a skill."

A user is frustrated with a slow or unmaintained tool they already have installed. Skill Finder searches for alternatives, filters out low-download or stale results, and recommends a better-maintained replacement with a side-by-side tradeoff comparison.

A team wants to standardize on Skills.sh for a particular workspace. The user sets the source mode to skills.sh, and Skill Finder saves that preference — all future searches in that workspace query only the Skills.sh ecosystem until changed.

A developer starting a new project asks "what should I install for React testing?" Skill Finder translates that into specific search queries, compares ClawHub and Skills.sh results, and surfaces the top 1-3 options with install commands and reasoning.