fix(scanner): comment-anchor TODO grep + skip test-file markers (Codex #2) (#15)

The TODO/FIXME/HACK/DOMAIN_QUESTION scan matched any occurrence of the
marker words anywhere on a line, so string literals, regexes, and the
scanner's own pattern produced phantom improvement tasks. It also surfaced
markers living in test fixtures as if they were real work.

- Anchor the grep pattern to a line-start comment opener
  (`^[ \t]*(//|\*|#)[ \t]*(TODO|FIXME|HACK|DOMAIN_QUESTION)[: ]`), so only
  genuine comment markers match — not bare string/regex occurrences.
- Post-filter results through a new exported `isTestFile()` helper to drop
  markers that live in `__tests__/` dirs or `*.test`/`*.spec` files.
- `GREP_EXCLUDE_DIRS` and `scanDeadCode` left unchanged on purpose: dead-code
  usage search still needs test files, so excluding them there would
  reintroduce false dead-code positives.

Adds regression tests: test-file markers ignored, the grep pattern stays
comment-anchored, and `isTestFile` classification.

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

Author: telivity-otaip

packages/asil-improvement-loop/src/__tests__/scanner.test.ts

@@ -1,6 +1,7 @@
 import { describe, expect, it } from 'vitest';
 import {
   normalizePath,
+  isTestFile,
   scanCodebase,
   scanTestFailures,
   scanTypeErrors,
@@ -172,6 +173,49 @@ describe('scanner', () => {
       const foo = tasks.find((t) => t.filePaths[0]?.includes('foo.ts'));
       expect(foo?.severity).toBe('medium');
     });
+
+    it('skips markers found in test files — they are fixtures, not tasks (precision fix)', async () => {
+      const grepOut = [
+        './packages/foo/src/__tests__/foo.test.ts:1:// DOMAIN_QUESTION: fixture data',
+        './packages/foo/src/scanner.spec.ts:2:// TODO: spec fixture',
+        './packages/foo/src/real.ts:9:// TODO: a genuine task',
+      ].join('\n');
+      const runner = mockRunner([{ match: () => true, stdout: grepOut }]);
+      const tasks = await scanTodos('/repo', { runner, fs: mockFileReader() });
+      // Only the non-test file yields a task.
+      expect(tasks.length).toBe(1);
+      expect(tasks[0]?.filePaths[0]).toBe('packages/foo/src/real.ts');
+    });
+
+    it('greps with a comment-anchored marker pattern (no bare string-literal matches)', async () => {
+      const seen: string[][] = [];
+      const runner = mockRunner([
+        {
+          match: (cmd, args) => {
+            if (cmd === 'grep') seen.push(args);
+            return cmd === 'grep';
+          },
+          stdout: '',
+        },
+      ]);
+      await scanTodos('/repo', { runner, fs: mockFileReader() });
+      const pattern = seen[0]?.find((a) => a.includes('DOMAIN_QUESTION')) ?? '';
+      // The pattern must require a line-start comment opener before the
+      // marker, so `'DOMAIN_QUESTION'` / `/DOMAIN_QUESTION:/` don't match.
+      expect(pattern).toContain('(//|\\*|#)');
+      expect(pattern.startsWith('^')).toBe(true);
+    });
+  });
+
+  describe('isTestFile', () => {
+    it('matches __tests__ dirs and .test/.spec files; not plain source', () => {
+      expect(isTestFile('packages/x/src/__tests__/a.ts')).toBe(true);
+      expect(isTestFile('src/a.test.ts')).toBe(true);
+      expect(isTestFile('src/a.spec.tsx')).toBe(true);
+      expect(isTestFile('a.test.py')).toBe(false); // py uses test_ prefix, not .test.
+      expect(isTestFile('src/a.ts')).toBe(false);
+      expect(isTestFile('src/scanner.ts')).toBe(false);
+    });
   });
 
   describe('scanCoverageGaps', () => {

packages/asil-improvement-loop/src/index.ts

@@ -21,6 +21,7 @@ export {
   scanCoverageGaps,
   scanDeadCode,
   normalizePath,
+  isTestFile,
   stableTaskId,
   toRepoRelative,
   type ScanResult,

packages/asil-improvement-loop/src/scanner.ts

@@ -157,7 +157,15 @@ export async function scanTodos(
       ...GREP_EXCLUDE_DIRS,
       ...includes,
       '-E',
-      '(TODO|FIXME|HACK|DOMAIN_QUESTION)[: ]',
+      // Require the marker to be the FIRST token of a line-start comment
+      // (`//`, `#`, or jsdoc `*`). This is ASIL's own DOMAIN_QUESTION
+      // convention and it eliminates the false positives a live grind
+      // surfaced: marker strings inside string literals / regexes
+      // (e.g. `'DOMAIN_QUESTION'`, `/DOMAIN_QUESTION:/`) and mid-comment
+      // mentions are no longer matched — only genuine actionable markers.
+      // Trade-off: trailing comments (`code; // TODO`) aren't matched,
+      // consistent with how DOMAIN_QUESTION markers are already detected.
+      '^[ \\t]*(//|\\*|#)[ \\t]*(TODO|FIXME|HACK|DOMAIN_QUESTION)[: ]',
       '.',
     ],
     { cwd: repoRoot },
@@ -173,6 +181,10 @@ export async function scanTodos(
     const [, file, ln, txt] = m;
     if (!file) continue;
     const key = normalizePath(file);
+    // Skip test files — a TODO/FIXME/DOMAIN_QUESTION marker in a test is
+    // almost always test DATA (fixtures exercising the scanner itself),
+    // not an actionable task. (Live-grind precision finding.)
+    if (isTestFile(key)) continue;
     const arr = byFile.get(key) ?? [];
     arr.push(`:${ln} — ${txt?.trim() ?? ''}`);
     byFile.set(key, arr);
@@ -341,6 +353,19 @@ function shortPath(p: string): string {
   return p.replace(/^.*\//, '');
 }
 
+/**
+ * True for files that are tests: anything under a `__tests__/` directory
+ * or named `*.test.*` / `*.spec.*`. Used to keep TODO/FIXME markers that
+ * are test FIXTURES (data exercising the scanner) out of the task queue.
+ */
+export function isTestFile(p: string): boolean {
+  return (
+    p.includes('/__tests__/') ||
+    p.startsWith('__tests__/') ||
+    /\.(test|spec)\.[cm]?[jt]sx?$/.test(p)
+  );
+}
+
 /**
  * Convert an absolute path under `repoRoot` to a repo-relative path.
  * Coverage tools (vitest's istanbul reporter, coverage.py) emit