Skip to main content

​
E-TOOLS-003

what doctor prints (warning)catalog: 5/12 tools ungraded — each one asks on every call; run vendo sync with a model key to grade
Some of your host tools have no risk grade, so the guard asks the user before every single call instead of letting reads through automatically. check tools/graded · error_code E-TOOLS-003 · doctor exits 0 (warning only)

​
What you’re seeing

Doctor counts how many tools in .vendo/tools.json have a grade once .vendo/judgments.json and .vendo/overrides.json are both applied, and warns when the ungraded count is above zero.

​
Why

Extraction only asserts protocol facts: an HTTP DELETE is destructive, a declared tRPC mutation is at least a write. It never guesses a grade from a tool’s name alone. Everything else stays ungraded until the AI judge or a human sets it by hand, and the guard treats ungraded as risky by default, asking before it runs.

​
The fix

Run vendo sync with a model key so the judgment pass grades the catalog against the real handler source:
Without a model key, grade the tools you care about directly:
Leaving tools ungraded is safe. It only costs an approval prompt per call.

E-TOOLS-001

every tool disabled

E-TOOLS-002

tool surface is empty

E-TOOLS-004

tools with no schema