SinachPat
6e59a8a36b
Stand up the monorepo skeleton and land the first Phase 0 artifacts: pairing threat model, plugin catalog, builder detect, 20-prompt golden harness, and synthetic 2GB mirror timing.
32 lines
1.0 KiB
TypeScript
32 lines
1.0 KiB
TypeScript
import { applyTool } from './apply-tool.ts';
|
|
import { checkAssertion } from './check-assertion.ts';
|
|
import type { GoldenAssert, GrokResponse, SiteFixture, ToolCall, ToolName } from './types.ts';
|
|
|
|
const tools = new Set<ToolName>(['edit_heading', 'edit_text', 'update_option']);
|
|
|
|
function isToolName(name: string): name is ToolName {
|
|
return tools.has(name as ToolName);
|
|
}
|
|
|
|
export function parseGrokToolCalls(grok: GrokResponse): ToolCall[] {
|
|
const calls = grok.choices[0]?.message.tool_calls ?? [];
|
|
return calls.map((call) => {
|
|
if (!isToolName(call.function.name)) {
|
|
throw new Error(`unknown tool: ${call.function.name}`);
|
|
}
|
|
return {
|
|
name: call.function.name,
|
|
arguments: JSON.parse(call.function.arguments) as Record<string, string>,
|
|
};
|
|
});
|
|
}
|
|
|
|
export function scoreGrokResponse(input: {
|
|
site: SiteFixture;
|
|
assert: GoldenAssert;
|
|
grok: GrokResponse;
|
|
}): { passed: boolean } {
|
|
const next = parseGrokToolCalls(input.grok).reduce(applyTool, input.site);
|
|
return { passed: checkAssertion(next, input.assert).ok };
|
|
}
|