@uinaf/skillcheck 1.1.0 → 1.2.0

package/dist/cli.js

@@ -80,13 +80,13 @@ function runOptions(flags) {
 	const judgeModel = flags.get("--judge") ?? "claude-opus-5";
 	const judgeEffort = flags.get("--judge-effort");
 	if (judgeEffort !== void 0) {
-		if (![
+		const levels = judgeModel.includes(":") ? [
 			"minimal",
 			"low",
 			"medium",
 			"high"
-		].includes(judgeEffort)) throw new Error(`--judge-effort must be minimal, low, medium, or high, got ${judgeEffort}`);
-		if (!judgeModel.includes(":")) throw new Error("--judge-effort needs a provider-qualified --judge (e.g. openai:chat:gpt-5.6-sol); the Anthropic judge does not take a reasoning effort");
+		] : AGENT_EFFORTS;
+		if (!levels.includes(judgeEffort)) throw new Error(`--judge-effort for ${judgeModel} must be ${levels.join(", ")}, got ${judgeEffort}`);
 	}
 	return {
 		harness,
@@ -304,7 +304,8 @@ function judgeName(judge) {
 	if (typeof judge === "string") return judge.replace(/^anthropic:messages:/, "");
 	const j = judge;
 	const name = j?.config?.model ?? j?.id;
-	return typeof name === "string" ? name : "unknown";
+	if (typeof name !== "string") return "unknown";
+	return j?.config?.effort === void 0 ? name : name.replace(/^anthropic:messages:/, "");
 }
 function resultRunConfig(raw, meta, harness) {
 	const m = meta;
@@ -318,7 +319,8 @@ function resultRunConfig(raw, meta, harness) {
 	const r = raw;
 	const agent = r?.config?.providers?.[0]?.config;
 	const judge = r?.config?.defaultTest?.options?.provider;
-	const judgeEffort = judge?.config?.reasoning_effort;
+	const judgeConfig = judge?.config;
+	const judgeEffort = judgeConfig?.reasoning_effort ?? judgeConfig?.effort;
 	return {
 		agent_model: typeof agent?.model === "string" ? agent.model : `${harness}-default`,
 		agent_effort: typeof agent?.effort === "string" ? agent.effort : null,

package/dist/scenario.js

@@ -179,10 +179,14 @@ function buildConfig(s, trials, opts, paths) {
 			provider: opts.judgeModel.includes(":") ? opts.judgeEffort === void 0 ? opts.judgeModel : {
 				id: opts.judgeModel,
 				config: { reasoning_effort: opts.judgeEffort }
-			} : process.env.ANTHROPIC_API_KEY ? `anthropic:messages:${opts.judgeModel}` : {
+			} : process.env.ANTHROPIC_API_KEY ? opts.judgeEffort === void 0 ? `anthropic:messages:${opts.judgeModel}` : {
+				id: `anthropic:messages:${opts.judgeModel}`,
+				config: { effort: opts.judgeEffort }
+			} : {
 				id: "anthropic:claude-agent-sdk",
 				config: {
 					model: opts.judgeModel,
+					...opts.judgeEffort ? { effort: opts.judgeEffort } : {},
 					apiKeyRequired: false,
 					max_turns: 3,
 					output_format: {

package/docs/usage.md

@@ -109,9 +109,16 @@ skillcheck run <scenario-dir> --judge openai:chat:gpt-5.6-sol --judge-effort hig
 
 A provider-qualified judge authenticates through that provider's own env
 (`OPENAI_API_KEY`, plus `OPENAI_BASE_URL` for a gateway) and is recorded
-verbatim in the scorecard's `judge_model` column. `--judge-effort`
-(minimal|low|medium|high) sets `reasoning_effort` and requires a
-provider-qualified judge; the Anthropic judge does not take one.
+verbatim in the scorecard's `judge_model` column. For it, `--judge-effort`
+(minimal|low|medium|high) sets `reasoning_effort`. For a bare Claude judge,
+`--judge-effort` takes Claude's levels (low|medium|high|xhigh|max) and is passed
+as `effort` on either Anthropic path; the SDK judge starts Claude Code with
+`--effort`:
+
+```sh
+skillcheck run <scenario-dir> --agent claude-opus-5-5 --agent-effort medium \
+  --judge claude-opus-5-5 --judge-effort high --trials 3
+```
 
 ## Sweep
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@uinaf/skillcheck",
-  "version": "1.1.0",
+  "version": "1.2.0",
   "description": "Lint and eval harness for agent skills",
   "homepage": "https://github.com/uinaf/skillcheck#readme",
   "bugs": {