Skills
Per-skill install pages: what the skill does, the install command verified against the author's own docs, and a trust review. A small, hand-checked directory — seventeen skills. Pages verified 2026-10-23 carry a trust score only where a full code read was done; the rest state plainly that no score is given.
- obra/superpowers — install for Claude CodeTrust score 7/10 · last verified 2026-10-19
- mattpocock/skills — install for CursorTrust score 6/10 · last verified 2026-10-19
- thedotmack/claude-mem — install for Claude CodeTrust score 5/10 · last verified 2026-10-19
- Leonxlnx/taste-skill — install for Claude CodeTrust score 6/10 · last verified 2026-10-19
- upstash/context7 — install for Claude CodeTrust score 5/10 · last verified 2026-10-19
- wshobson/agents — install for Claude CodeNo score — full code read not completed · last verified 2026-10-23
- vercel-labs/agent-skills — install for Claude CodeNo score — full code read not completed · last verified 2026-10-23
- anthropics/skills — install for Claude CodeNo score — full code read not completed · last verified 2026-10-23
- microsoft/playwright-mcp — install for Claude CodeNo score — full code read not completed · last verified 2026-10-23
- supabase-community/supabase-mcp — install for MCP clientsNo score — full code read not completed · last verified 2026-10-23
- exa-labs/exa-mcp-server — install for Claude CodeTrust score 7/10 (full code read) · last verified 2026-10-23
- getsentry/sentry-mcp — install for Claude CodeNo score — full code read not completed · last verified 2026-10-23
- stripe/agent-toolkit — install for Claude CodeNo score — full code read not completed · last verified 2026-10-23
- github/awesome-copilot — install for GitHub CopilotNo score — full code read not completed · last verified 2026-10-23
- qdrant/mcp-server-qdrant — install for Claude DesktopNo score — full code read not completed · last verified 2026-10-23
- neondatabase/mcp-server-neon — install for Claude CodeNo score — full code read not completed · last verified 2026-10-23
- modelcontextprotocol/servers — install for Claude DesktopNo score — full code read not completed · last verified 2026-10-23
How the trust score works
Every skill page carries a score out of 10 against the same three checks:
- Skill-file review — what the skill's manifest and instruction files actually tell the agent to do. Anything the skill asks for beyond its stated job costs points.
- Install behavior — what happens when you install: does it run code, fetch remote payloads, or phone home? Read-only file installs score highest.
- Prompt-injection surface — how exposed the skill is to being steered by untrusted content: skill instructions that read untrusted input, memory layers, and retrieval paths that mix sources all cost points.
Scores are a public-code review as of the last-verified date — a signal, not a certification. We re-check when the skill changes. When in doubt, we say what we verified and what we didn't.