You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
#3654 added NameValidation (MCP v2025-11-25 tool-name rules) and the code comment above it says "we'll be implementing it across Toolbox", but as of main it is only wired into UnmarshalYAMLToolConfig (internal/server/config.go:387). Sources, authServices, prompts, and embeddingModels still accept any name.
Question
Which kinds should get the same validation, and is there a breaking-change window to respect?
Prompts seem the clearest next step: prompt names go over the MCP wire via prompts/list and prompts/get, the same protocol surface as tool names, so the spec's name rules arguably apply there too.
Sources / authServices are toolbox-internal references (a tool's source: field points at one). Names with spaces or commas currently parse fine, so tightening them would break existing configs that fix: add name validation to tools name #3654 deliberately left alone. If these should be validated too, a deprecation path might be needed rather than an outright rejection.
Happy to open a PR for whatever scope you prefer (prompts only, or prompts + sources with some compatibility story). Asking first since the right cut is a spec/policy call rather than a technical one.
Background
#3654 added
NameValidation(MCP v2025-11-25 tool-name rules) and the code comment above it says "we'll be implementing it across Toolbox", but as ofmainit is only wired intoUnmarshalYAMLToolConfig(internal/server/config.go:387). Sources, authServices, prompts, and embeddingModels still accept any name.Question
Which kinds should get the same validation, and is there a breaking-change window to respect?
prompts/listandprompts/get, the same protocol surface as tool names, so the spec's name rules arguably apply there too.source:field points at one). Names with spaces or commas currently parse fine, so tightening them would break existing configs that fix: add name validation to tools name #3654 deliberately left alone. If these should be validated too, a deprecation path might be needed rather than an outright rejection.Happy to open a PR for whatever scope you prefer (prompts only, or prompts + sources with some compatibility story). Asking first since the right cut is a spec/policy call rather than a technical one.