@uinaf/skillcheck 1.0.0 → 1.0.1

package/dist/cli.js

@@ -235,7 +235,7 @@ function cmdSweep(argv) {
 	for (const dir of discoverScenarios(root)) {
 		const name = runNameFor(dir, opts.harness);
 		const resultPath = path.join(resultsDir, `${name}.json`);
-		if (!all && fs.existsSync(resultPath)) {
+		if (!all && fs.existsSync(resultPath) && !fs.existsSync(attemptPath(resultPath))) {
 			skipped++;
 			console.log(`SKIP  ${name} (results exist; use --all to rerun)`);
 			continue;
@@ -256,13 +256,21 @@ function cmdSweep(argv) {
 	process.exit(errored > 0 ? 2 : failed > 0 ? 1 : 0);
 }
 function resultIdentity(file) {
+	const decode = (part) => {
+		if (!part.startsWith("~v2~")) return part;
+		try {
+			return decodeURIComponent(part.slice(4));
+		} catch {
+			return part;
+		}
+	};
 	const base = file.replace(/\.json$/, "");
 	const suffix = base.match(/--(codex|grok|cursor)$/);
 	const harness = suffix === null ? "claude" : suffix[1];
 	const [skill, ...rest] = base.replace(/--(codex|grok|cursor)$/, "").split("--");
 	return {
-		skill,
-		scenario: rest.join("--"),
+		skill: decode(skill),
+		scenario: decode(rest.join("--")),
 		harness
 	};
 }

package/dist/grok-provider.js

@@ -113,6 +113,11 @@ var GrokProvider = class {
 			const abort = () => stop("grok run aborted");
 			callOptions?.abortSignal?.addEventListener("abort", abort, { once: true });
 			if (callOptions?.abortSignal?.aborted) abort();
+			child.on("exit", () => {
+				if (process.platform !== "win32" && child.pid !== void 0) try {
+					process.kill(-child.pid, "SIGKILL");
+				} catch {}
+			});
 			child.on("close", (code, signal) => {
 				clearTimeout(timer);
 				callOptions?.abortSignal?.removeEventListener("abort", abort);

package/dist/scenario.js

@@ -32,10 +32,15 @@ function loadScenario(scenarioDir) {
 		criteria
 	};
 }
+function encodeRunNamePart(part) {
+	if (!part.includes("--") && !part.startsWith("-") && !part.endsWith("-") && !part.startsWith("~v2~")) return part;
+	return `~v2~${encodeURIComponent(part).replaceAll("-", "%2D").replaceAll("~", "%7E")}`;
+}
 function runNameFor(scenarioDir, harness) {
 	const m = path.resolve(scenarioDir).match(/skills\/([^/]+)\/evals\/([^/]+)$/);
 	if (!m) throw new Error(`not a scenario dir (want .../skills/<skill>/evals/<scenario>): ${scenarioDir}`);
-	return harness === "claude" ? `${m[1]}--${m[2]}` : `${m[1]}--${m[2]}--${harness}`;
+	const name = `${encodeRunNamePart(m[1])}--${encodeRunNamePart(m[2])}`;
+	return harness === "claude" ? name : `${name}--${harness}`;
 }
 function frontmatterRange(text) {
 	const lines = text.split("\n");
@@ -258,4 +263,4 @@ function generateRun(scenarioDir, opts, paths) {
 	};
 }
 //#endregion
-export { buildConfig, generateRun, isHiddenSkill, loadScenario, materialize, requiredEvalPackages, resolvePackageDir, runNameFor, sdkNodeModulesDir, stripHiddenFlag };
+export { buildConfig, encodeRunNamePart, generateRun, isHiddenSkill, loadScenario, materialize, requiredEvalPackages, resolvePackageDir, runNameFor, sdkNodeModulesDir, stripHiddenFlag };

package/docs/scenarios.md

@@ -9,7 +9,9 @@ A scenario is two files in a frozen location:
 
 The path is the identity: `<skill>--<scenario>` names the run, the result file,
 and the scorecard entry. Codex and Grok results gain `--codex` and `--grok`
-suffixes, so harnesses can hold results side by side.
+suffixes, so harnesses can hold results side by side. Double hyphens and
+leading or trailing hyphens in a skill or scenario name are escaped in the run
+and result filename; the scorecard keeps the original name.
 A directory missing either file is not discovered.
 
 ## task.md

package/docs/usage.md

@@ -77,7 +77,7 @@ provider-qualified judge; the Anthropic judge does not take one.
 ## Sweep
 
 ```sh
-skillcheck sweep           # only scenarios without results
+skillcheck sweep           # scenarios without completed results
 skillcheck sweep --all     # rerun everything
 ```
 
@@ -91,8 +91,8 @@ within one scenario, not across them.
 One known failure mode: judge calls through a gateway can drop at the transport
 layer ([uinaf/zebroid-infra#44](https://github.com/uinaf/zebroid-infra/issues/44)).
 That surfaces as an ERROR with no usable result, not as a graded FAIL, and the
-mitigation is a rerun. `sweep` without `--all` resumes, so a rerun only picks up
-what is missing.
+mitigation is a rerun. `sweep` without `--all` resumes, so a rerun picks up
+missing or ungraded results.
 
 ## Summarize
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@uinaf/skillcheck",
-  "version": "1.0.0",
+  "version": "1.0.1",
   "description": "Lint and eval harness for agent skills",
   "homepage": "https://github.com/uinaf/skillcheck#readme",
   "bugs": {