@uinaf/skillcheck 1.5.1 → 1.5.3

package/dist/scenario.js

@@ -1,3 +1,4 @@
+import { SKIPPED_DIRS } from "./transform.js";
 import { createRequire } from "node:module";
 import { createHash } from "node:crypto";
 import fs from "node:fs";
@@ -74,6 +75,41 @@ function stripHiddenFlag(skillMd) {
 	if (!range) return skillMd;
 	return skillMd.split("\n").filter((l, i) => !(i >= range[0] && i < range[1] && l.startsWith("disable-model-invocation:"))).join("\n");
 }
+const MARKDOWN_LINK = /\]\(\s*<([^>]+)>|\]\(\s*([^)\s]+)|^ {0,3}\[[^\]]+\]:\s*<?([^\s>]+)/gm;
+function linkedSiblings(skillDir) {
+	const skillsRoot = path.dirname(skillDir);
+	const realRoot = fs.realpathSync(skillsRoot);
+	const self = path.basename(skillDir);
+	const found = /* @__PURE__ */ new Set([self]);
+	const pending = [skillDir];
+	for (let dir = pending.pop(); dir !== void 0; dir = pending.pop()) for (const file of skillMarkdown(dir)) for (const m of fs.readFileSync(file, "utf8").matchAll(MARKDOWN_LINK)) {
+		const href = m[1] ?? m[2] ?? m[3];
+		if (href.startsWith("#") || /^[a-z][a-z0-9+.-]*:/i.test(href)) continue;
+		const target = path.resolve(path.dirname(file), href.split("#")[0]);
+		const rel = path.relative(skillsRoot, target);
+		const name = rel.split(path.sep)[0];
+		if (name === "" || name === ".." || path.isAbsolute(rel) || found.has(name)) continue;
+		if (!fs.existsSync(path.join(skillsRoot, name, "SKILL.md"))) continue;
+		if (path.dirname(fs.realpathSync(path.join(skillsRoot, name))) !== realRoot) continue;
+		found.add(name);
+		pending.push(path.join(skillsRoot, name));
+	}
+	found.delete(self);
+	return [...found].sort();
+}
+function skillMarkdown(dir) {
+	const out = [];
+	const walk = (d) => {
+		for (const e of fs.readdirSync(d, { withFileTypes: true })) {
+			const p = path.join(d, e.name);
+			if (e.isDirectory()) {
+				if (!(d === dir && e.name === "evals")) walk(p);
+			} else if (e.isFile() && e.name.endsWith(".md")) out.push(p);
+		}
+	};
+	walk(dir);
+	return out;
+}
 const RESERVED = /* @__PURE__ */ new Set([
 	".claude",
 	".agents",
@@ -105,10 +141,15 @@ function materialize(s, runDir, harness, control = false) {
 		fs.writeFileSync(dest, content);
 	}
 	const roots = control ? [] : harness === "codex" ? [".claude", ".agents"] : harness === "grok" ? [".grok"] : [".claude"];
-	for (const root of roots) fs.cpSync(s.skillDir, path.join(workdir, root, "skills", s.skill), {
-		recursive: true,
-		filter: (src) => path.basename(src) !== "evals"
-	});
+	const skillsRoot = path.dirname(s.skillDir);
+	const installed = control ? [] : [s.skill, ...linkedSiblings(s.skillDir)];
+	for (const root of roots) for (const name of installed) {
+		const from = path.join(skillsRoot, name);
+		fs.cpSync(from, path.join(workdir, root, "skills", name), {
+			recursive: true,
+			filter: (src) => src !== path.join(from, "evals")
+		});
+	}
 	for (const root of roots) {
 		const skillMd = path.join(workdir, root, "skills", s.skill, "SKILL.md");
 		const text = fs.readFileSync(skillMd, "utf8");
@@ -120,7 +161,7 @@ function materialize(s, runDir, harness, control = false) {
 		for (const e of fs.readdirSync(dir, { withFileTypes: true })) {
 			const p = path.join(dir, e.name);
 			if (e.isDirectory()) {
-				if (e.name !== ".claude" && e.name !== ".grok" && e.name !== "node_modules") walk(p);
+				if (!SKIPPED_DIRS.has(e.name)) walk(p);
 			} else manifest[path.relative(workdir, p)] = createHash("sha256").update(fs.readFileSync(p)).digest("hex");
 		}
 	};
@@ -151,7 +192,11 @@ function agentProvider(opts, workdir, skill, paths) {
 			sandbox_mode: "workspace-write",
 			network_access_enabled: true,
 			web_search_enabled: true,
-			cli_env: { CODEX_HOME: path.join(workdir, "..", "..", "codex-home") }
+			cli_config: { features: { plugins: false } },
+			cli_env: {
+				CODEX_HOME: path.join(workdir, "..", "..", "codex-home"),
+				HOME: path.join(workdir, "..", "..", "home")
+			}
 		}
 	};
 	return {
@@ -293,6 +338,26 @@ function privateCodexHome(dir) {
 		if (fs.existsSync(from)) fs.symlinkSync(from, path.join(dir, file));
 	}
 }
+const HIDDEN_HOME_ENTRIES = /* @__PURE__ */ new Set([
+	".agents",
+	".claude",
+	".codex"
+]);
+function privateHome(dir) {
+	const source = os.homedir();
+	fs.rmSync(dir, {
+		recursive: true,
+		force: true
+	});
+	fs.mkdirSync(dir, {
+		recursive: true,
+		mode: 448
+	});
+	for (const name of fs.readdirSync(source)) {
+		if (HIDDEN_HOME_ENTRIES.has(name)) continue;
+		fs.symlinkSync(path.join(source, name), path.join(dir, name));
+	}
+}
 function generateRun(scenarioDir, opts, paths) {
 	const s = loadScenario(scenarioDir);
 	const name = runNameFor(scenarioDir, opts.harness, opts.control);
@@ -302,7 +367,10 @@ function generateRun(scenarioDir, opts, paths) {
 		force: true
 	});
 	const trials = Array.from({ length: opts.trials ?? 1 }, (_, i) => materialize(s, path.join(runDir, trialLabel(i)), opts.harness, opts.control));
-	if (opts.harness === "codex") privateCodexHome(path.join(runDir, "codex-home"));
+	if (opts.harness === "codex") {
+		privateCodexHome(path.join(runDir, "codex-home"));
+		privateHome(path.join(runDir, "home"));
+	}
 	const sdkDir = sdkNodeModulesDir();
 	if (sdkDir !== void 0) {
 		const link = path.join(runDir, "node_modules");
@@ -323,4 +391,4 @@ function generateRun(scenarioDir, opts, paths) {
 	};
 }
 //#endregion
-export { DEFAULT_CLAUDE_AGENT, FILE_BLOCK, buildConfig, encodeRunNamePart, generateRun, isHiddenSkill, loadScenario, materialize, privateCodexHome, requiredEvalPackages, resolvePackageDir, runNameFor, sdkNodeModulesDir, stripHiddenFlag, trialLabel };
+export { DEFAULT_CLAUDE_AGENT, FILE_BLOCK, buildConfig, encodeRunNamePart, generateRun, isHiddenSkill, linkedSiblings, loadScenario, materialize, privateCodexHome, privateHome, requiredEvalPackages, resolvePackageDir, runNameFor, sdkNodeModulesDir, stripHiddenFlag, trialLabel };

package/dist/transform.js

@@ -14,6 +14,14 @@ function capped(rel, text) {
 	if (text.length <= PER_FILE_CAP) return `=== OUTPUT FILE: ${rel} ===\n${text}\n=== END OUTPUT FILE ===`;
 	return `=== OUTPUT FILE: ${rel} (truncated: showing ${PER_FILE_CAP} of ${text.length} chars) ===\n${safeSlice(text, PER_FILE_CAP)}\n=== END OUTPUT FILE ===`;
 }
+const SKIPPED_DIRS = /* @__PURE__ */ new Set([
+	".claude",
+	".grok",
+	"node_modules",
+	".cache",
+	".npm",
+	".pnpm-store"
+]);
 function transform(output, context) {
 	const workdir = context.vars.workdir;
 	const manifest = JSON.parse(fs.readFileSync(context.vars.manifest, "utf8"));
@@ -23,7 +31,7 @@ function transform(output, context) {
 		for (const e of fs.readdirSync(dir, { withFileTypes: true })) {
 			const p = path.join(dir, e.name);
 			if (e.isDirectory()) {
-				if (e.name !== ".claude" && e.name !== ".grok" && e.name !== "node_modules") walk(p);
+				if (!SKIPPED_DIRS.has(e.name)) walk(p);
 			} else if (e.isFile()) {
 				const rel = path.relative(workdir, p);
 				visited.add(rel);
@@ -64,4 +72,4 @@ function transform(output, context) {
 	return `${output}\n\n${appended}`;
 }
 //#endregion
-export { transform as default };
+export { SKIPPED_DIRS, transform as default };

package/docs/scenarios.md

@@ -114,7 +114,19 @@ time. It stays outside the root so the agent cannot reach the skill's source,
 its evals, or the repository's own agent guidance. The skill under test is installed where the harness discovers skills:
 `.claude/skills/<skill>/`, plus `.agents/skills/<skill>/` on codex, or
 `.grok/skills/<skill>/` on Grok, with its `evals/` directory excluded, so
-criteria never leak into the agent's context.
+criteria never leak into the agent's context. Sibling skills under the same root that the skill links to (inline, angle-bracket, or reference links, followed transitively; a symlink out of the root is skipped) are installed beside it
+the same way, so links such as `../other/SKILL.md` resolve as they do when a
+plugin installs the set; a control run installs none of them.
+
+A Codex run gets a private `CODEX_HOME` holding only the operator's
+`config.toml` and `auth.json`, with plugins turned off (`features.plugins =
+false`) so a plugin the config enables cannot supply skills, and a private `HOME` that links every entry of
+the operator's home except `.agents`, `.claude`, and `.codex`. Auth helpers that read
+`$HOME` keep working; the operator's user-level skills and global guidance are
+not visible.
+
+Graded output skips agent config roots, `node_modules`, and tool caches
+(`.cache`, `.npm`, `.pnpm-store`), so a cache cannot crowd deliverables out.
 
 Scenario quality is behavioral proof; [authoring](authoring.md) covers the
 judgment layer lint and evals cannot grade.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@uinaf/skillcheck",
-  "version": "1.5.1",
+  "version": "1.5.3",
   "description": "Lint and eval harness for agent skills",
   "homepage": "https://github.com/uinaf/skillcheck#readme",
   "bugs": {
@@ -66,5 +66,5 @@
   "engines": {
     "node": ">=24"
   },
-  "packageManager": "pnpm@12.5.1"
+  "packageManager": "pnpm@12.6.0"
 }