← ClaudeAtlas

tool-vettinglisted

Use when evaluating external MCP servers, APIs, SaaS tools, connectors, or paid services before PM Copilot depends on them.
Leo-maomao/pm-copilot · ★ 4 · AI & Automation · score 77
Install: claude install-skill Leo-maomao/pm-copilot
# Tool Vetting ## Goal Prevent PM Copilot from relying on unavailable, unverified, unsafe, or unexpectedly paid tools. A tool is usable only when its source, setup path, authentication, cost risk, data access, permissions, and fallback are explicit. ## Workflow 1. Define the job the tool must do and whether the job is required or optional. 2. Check `tools/external-tool-catalog.json` first. If the tool is absent, treat it as unvetted until source verification is completed. 3. Prefer official providers and mature open-source projects. Community wrappers can be used only with a visible maintenance and fallback note. 4. Verify the source before recommending the tool: official docs, official repo, package page, or inspected community repo. 5. Classify runtime status separately from candidate quality: `candidate` means promising but not configured, `setup_required` means an install/API key/OAuth/account/browser/permission is missing, `available` means locally configured or supplied by the active agent runtime, and `blocked` means the source, credentials, permission scope, cost, or account risk makes it unusable. 6. Classify cost risk: `free_local`, `open_source_self_hostable`, `byo_account`, `commercial_api`, `unknown`, or `paid_only`. 7. Classify data risk: public-only, workspace documents, product analytics, production database, customer data, CRM/support data, advertising spend, or write-capable operations. 8. Require least-privilege credentials. Prefer read-only scopes for