Add skill governance and runtime routing
This commit is contained in:
1 parent
6f09408423
commit
41dcfcdef7
71 files changed
+7057
-102
No files matched your search
@@ -6,6 +6,8 @@ import { createBuiltInToolRegistry } from "@tools/builtins/index";
|
||||
import type { ProcessTraceSnapshot } from "@skills/onboarding/runtime/trace";
|
||||
import {
|
||||
getEvalProcessLlmJudgeMaxTokens,
|
||||
getEvalPrmEnabled,
|
||||
getEvalPrmWeightsPath,
|
||||
getLlmJudgeModel,
|
||||
} from "../../config.js";
|
||||
import type {
|
||||
@@ -13,6 +15,12 @@ import type {
|
||||
ProcessJudgeCheck,
|
||||
ProcessJudgeResult,
|
||||
} from "../types.js";
|
||||
import { attributeTraceWithPrm } from "./attribution.js";
|
||||
import {
|
||||
extractPrmFeatures,
|
||||
loadPrmModel,
|
||||
predictPrmScore,
|
||||
} from "./prm.js";
|
||||
|
||||
const processLlmJudgeSchema = z.object({
|
||||
rationale: z.string().trim().min(1).max(300),
|
||||
@@ -70,6 +78,37 @@ function finalize(checks: ProcessJudgeCheck[], rationale?: string): ProcessJudge
|
||||
return result;
|
||||
}
|
||||
|
||||
function enrichWithPrm(
|
||||
result: ProcessJudgeResult,
|
||||
trace: ProcessTraceSnapshot,
|
||||
): ProcessJudgeResult {
|
||||
if (!getEvalPrmEnabled()) {
|
||||
return result;
|
||||
}
|
||||
|
||||
const prmFeatures = extractPrmFeatures(trace);
|
||||
const model = loadPrmModel(getEvalPrmWeightsPath());
|
||||
if (!model) {
|
||||
return {
|
||||
...result,
|
||||
prmFeatures,
|
||||
};
|
||||
}
|
||||
|
||||
const attribution = attributeTraceWithPrm(trace, model);
|
||||
return {
|
||||
...result,
|
||||
prmFeatures,
|
||||
prmScore: predictPrmScore(model, prmFeatures),
|
||||
...(attribution.rootCauseStep
|
||||
? {
|
||||
rootCauseScoreLoss: attribution.rootCauseScoreLoss,
|
||||
rootCauseStep: attribution.rootCauseStep,
|
||||
}
|
||||
: {}),
|
||||
};
|
||||
}
|
||||
|
||||
function hasAllTools(
|
||||
actualTools: string[],
|
||||
expectedTools: string[],
|
||||
@@ -207,6 +246,23 @@ export function judgeProcessRule(
|
||||
|
||||
checks.push(validateToolArgs(trace));
|
||||
|
||||
const toolPlanStepCount = trace.steps.plan?.output.toolPlanStepCount;
|
||||
if (toolPlanStepCount !== undefined) {
|
||||
const actualToolCalls = trace.steps.toolCalls?.length ?? 0;
|
||||
checks.push(
|
||||
makeCheck(
|
||||
"tool-call",
|
||||
"tool-plan-consistent",
|
||||
actualToolCalls >= toolPlanStepCount,
|
||||
`planned=${toolPlanStepCount}, actual=${actualToolCalls}`,
|
||||
),
|
||||
);
|
||||
} else {
|
||||
checks.push(
|
||||
makeSkippedCheck("tool-call", "tool-plan-consistent", "no toolPlan"),
|
||||
);
|
||||
}
|
||||
|
||||
if (processExpected?.minSteps !== undefined) {
|
||||
const actualSteps = countRecordedSteps(trace);
|
||||
checks.push(
|
||||
@@ -258,7 +314,7 @@ export function judgeProcessRule(
|
||||
);
|
||||
}
|
||||
|
||||
return finalize(checks);
|
||||
return enrichWithPrm(finalize(checks), trace);
|
||||
}
|
||||
|
||||
function buildProcessJudgePrompt(
|
||||
@@ -289,7 +345,8 @@ export async function judgeProcessWithLlm(
|
||||
options: { finalText: string; request: AgentRequest },
|
||||
): Promise<ProcessJudgeResult> {
|
||||
if (!llmClient.hasEndpoints()) {
|
||||
return finalize(
|
||||
return enrichWithPrm(
|
||||
finalize(
|
||||
[
|
||||
makeSkippedCheck(
|
||||
"trace",
|
||||
@@ -298,6 +355,8 @@ export async function judgeProcessWithLlm(
|
||||
),
|
||||
],
|
||||
"skipped: no LLM endpoint",
|
||||
),
|
||||
trace,
|
||||
);
|
||||
}
|
||||
|
||||
@@ -324,30 +383,39 @@ export async function judgeProcessWithLlm(
|
||||
);
|
||||
|
||||
if (!response) {
|
||||
return finalize(
|
||||
[makeCheck("trace", "validate-no-reject", false, "judge call failed")],
|
||||
"process judge call failed",
|
||||
return enrichWithPrm(
|
||||
finalize(
|
||||
[makeCheck("trace", "validate-no-reject", false, "judge call failed")],
|
||||
"process judge call failed",
|
||||
),
|
||||
trace,
|
||||
);
|
||||
}
|
||||
|
||||
try {
|
||||
const parsed = processLlmJudgeSchema.parse(JSON.parse(response.content));
|
||||
const passed = parsed.score >= 0.7;
|
||||
return finalize(
|
||||
[
|
||||
makeCheck(
|
||||
"trace",
|
||||
"validate-no-reject",
|
||||
passed,
|
||||
`llmScore=${round(parsed.score)}`,
|
||||
),
|
||||
],
|
||||
parsed.rationale,
|
||||
return enrichWithPrm(
|
||||
finalize(
|
||||
[
|
||||
makeCheck(
|
||||
"trace",
|
||||
"validate-no-reject",
|
||||
passed,
|
||||
`llmScore=${round(parsed.score)}`,
|
||||
),
|
||||
],
|
||||
parsed.rationale,
|
||||
),
|
||||
trace,
|
||||
);
|
||||
} catch {
|
||||
return finalize(
|
||||
[makeCheck("trace", "validate-no-reject", false, "invalid judge json")],
|
||||
"invalid process judge json",
|
||||
return enrichWithPrm(
|
||||
finalize(
|
||||
[makeCheck("trace", "validate-no-reject", false, "invalid judge json")],
|
||||
"invalid process judge json",
|
||||
),
|
||||
trace,
|
||||
);
|
||||
}
|
||||
}
|
||||
Reference in new issue
Block a user