vigiles 21.0.2 → 23.0.0

package/dist/test-coverage.js

@@ -273,7 +273,14 @@ function discoverTests(basePath, globs, ignore) {
     // skill (matched by the explicit-dot `.claude/skills/*/SKILL.md` pattern) is
     // still discovered — so the surface looks untested even after the user adds
     // exactly the suggested file. DEFAULT_IGNORE still drops .git/node_modules/etc.
-    const found = (0, glob_1.globSync)([...globs], { cwd: basePath, ignore, dot: true });
+    // `{surface}` is widened to `*` for DISCOVERY so one pass finds every
+    // candidate; the narrowing back to a specific surface happens at match time
+    // (matchesSurfaceGlob). Globbing once per surface would be quadratic in I/O.
+    const found = (0, glob_1.globSync)(globs.map(coverage_evidence_js_1.discoveryGlob), {
+        cwd: basePath,
+        ignore,
+        dot: true,
+    });
     // Prepared ONCE per file (comment-strip + declaration parse), not once per
     // (surface × file) pair — the matching below is quadratic by nature.
     return found.map((path) => (0, coverage_evidence_js_1.prepareTest)(path));
@@ -318,15 +325,17 @@ function isColocated(surface, testPath) {
  * explicitly declared is reported as declared; otherwise the provenance summary
  * would depend on glob order.
  */
-function coverageOf(surface, tests) {
+function coverageOf(surface, tests, globs) {
     let best = null;
     for (const t of tests) {
         if (t.path === surface.path)
             continue;
-        const ev = (0, coverage_evidence_js_1.evidenceFor)(surface, t, isColocated(surface, t.path));
+        const ev = (0, coverage_evidence_js_1.evidenceFor)(surface, t, isColocated(surface, t.path), (0, coverage_evidence_js_1.matchesSurfaceGlob)(surface, t.path, globs));
         if (!ev)
             continue;
-        if (!best)
+        // Rank, do not first-win: with a colocated harness AND a configured suite
+        // the reported provenance must not depend on glob order.
+        if (!best || (0, coverage_evidence_js_1.strongerEvidence)(ev, best.evidence))
             best = { surface, evidence: ev, by: t.path };
     }
     return best;
@@ -370,12 +379,12 @@ function executedOf(surface, index, tier) {
  * first, a colocated test second. Execution outranks the name because the name
  * was only ever a stand-in for it.
  */
-function tierOf(considered, tests, index, tier) {
+function tierOf(considered, tests, index, tier, globs) {
     const covered = [];
     const untested = [];
     const decisions = [];
     for (const s of considered) {
-        const decision = executedOf(s, index, tier) ?? coverageOf(s, tests);
+        const decision = executedOf(s, index, tier) ?? coverageOf(s, tests, globs);
         if (decision) {
             covered.push(s);
             decisions.push(decision);
@@ -432,7 +441,7 @@ function findUntestedSurfaces(options = {}) {
     // file has several legitimate spellings (`x.mjs`, `./x.mjs`, absolute), and
     // the artifact records whichever one was typed.
     (by) => (0, node_fs_1.existsSync)((0, node_path_1.join)(basePath, (0, coverage_artifact_js_1.canonicalScript)(by, basePath))));
-    const union = tierOf(considered, tests, runIndex, undefined);
+    const union = tierOf(considered, tests, runIndex, undefined, globs);
     return {
         total: considered.length,
         covered: union.covered,
@@ -456,8 +465,8 @@ function findUntestedSurfaces(options = {}) {
             .filter((path) => read((0, node_path_1.join)(basePath, path)).includes(LEGACY_COVERS)),
         retiredTestNames: retiredTestNamesFor(basePath, union.untested),
         decisions: union.decisions,
-        harness: tierOf(considered, split.harness, runIndex, "harness"),
-        evals: tierOf(considered, split.evals, runIndex, "eval"),
+        harness: tierOf(considered, split.harness, runIndex, "harness", globs),
+        evals: tierOf(considered, split.evals, runIndex, "eval", globs),
     };
 }
 /**
@@ -813,13 +822,17 @@ function formatUntestedReport(report) {
         lines.push(`  ${provenance}`);
     lines.push(...coverageCaveats(report));
     // Already testing these another way (a promptfoo suite, a home-grown evals
-    // file)? Point `testGlobs` at it so it counts toward coverage (issue #113) —
-    // and put the file NEXT TO the surface, which is the only placement that
-    // counts now. See docs/rules/untested-skill.md.
-    lines.push(`  Testing these another way (promptfoo / a home-grown eval loop)? Add its ` +
-        `files to \`testGlobs\` in .vigilesrc.json AND name each after the surface ` +
-        `it covers, next to it (\`<surface>/<surface>.eval.mjs\`) — placement says ` +
-        `where a file sits, the name says what it is about. ` +
+    // file)? Two shapes are accepted, and the message names BOTH — it used to
+    // name only `testGlobs`, which does not by itself make a centralized suite
+    // count, so a reader who followed it exactly saw the number not move (#175.2).
+    // See docs/rules/untested-skill.md.
+    lines.push(`  Testing these another way (promptfoo / a home-grown eval loop)? Either ` +
+        `put the file NEXT TO the surface and name it after it ` +
+        `(\`<surface>/<surface>.eval.mjs\`), or — for a centralized layout — point ` +
+        `\`testGlobs\` at it USING THE \`{surface}\` placeholder, e.g. ` +
+        `\`"tests/{surface}/evals/promptfooconfig*.yaml"\`. A \`testGlobs\` entry ` +
+        `WITHOUT \`{surface}\` widens what counts as a test file but never says ` +
+        `which surface it covers, so it credits nothing on its own. ` +
         `See docs/rules/untested-skill.md.`);
     return lines.join("\n");
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "vigiles",
-  "version": "21.0.2",
+  "version": "23.0.0",
   "description": "Audit, test and measure the harness your AI agent runs on — grade your CLAUDE.md / AGENTS.md, skills, subagents and hooks, run them against a scripted model, and measure whether they actually fire.",
   "keywords": [
     "claude-code",