fix(runner): alias params/input to args so tool implementations don't ReferenceError
Some checks failed
Deploy to Production / deploy (push) Has been cancelled
Some checks failed
Deploy to Production / deploy (push) Has been cancelled
Auth chain finally landed but tool calls crashed in the wetter server
with "Error: params is not defined". The MCP SDK passes the validated
tool args as a single parameter; our template names that parameter
`args` but the model frequently writes `params.location` / `input.x`
because that's how OpenAPI and JSON-RPC reference docs read.
Two-sided fix:
- render.ts wraps every implementation with `const params = args; const
input = args;` inside the try block. Whichever alias the model
picked, the variable resolves to the same validated object.
- SYSTEM_PROMPT now states the variable name EXPLICITLY ("variable
named EXACTLY `args`, e.g. args.location") so new generations stop
drifting on that detail.
Existing wetter runner needs a rebuild to pick up the alias shim.
This commit is contained in:
@@ -38,6 +38,15 @@ function renderTool(tool: ToolSpec): string {
|
||||
inputSchema: ${schemaShape},
|
||||
},
|
||||
async (args) => {
|
||||
// The MCP SDK passes the validated tool arguments as the single
|
||||
// parameter. Models trained on OpenAPI / JSON-RPC examples reach
|
||||
// for "params" instead of "args", and "input" shows up too — bind
|
||||
// every common alias to the same object so the generated body
|
||||
// works whichever name the model picked. Without this the runner
|
||||
// crashes with "ReferenceError: params is not defined" at the
|
||||
// first tool call (verified in prod with the wetter server).
|
||||
const params = args;
|
||||
const input = args;
|
||||
try {
|
||||
${tool.implementation}
|
||||
} catch (err) {
|
||||
|
||||
Reference in New Issue
Block a user