claudemd-cli 0.9.19 → 0.9.21

package/CHANGELOG.md

@@ -8,6 +8,46 @@ All notable changes to the `claudemd` plugin. This changelog tracks plugin artif
 - **Canonical spec version source**: `spec/CLAUDE.md` top-line title (`# AI-CODING-SPEC vX.Y.Z — Core`) + `spec/CLAUDE-changelog.md` top `##` entry.
 - **Plugin semver vs spec semver** are independent: plugin patch (0.2.0 → 0.2.1) may ship when spec is unchanged (this release); plugin minor (0.1.9 → 0.2.0) ships when spec minor updates (v0.2.0 shipped spec v6.10.0).
 
+## [0.9.21] - 2026-05-10
+
+**Hotfix — `claudemd-cli lint <path>` silent-text-scan when path looks like a path but doesn't resolve to a regular file (v0.9.14 family residual).** Spec v6.11.7 unchanged. The v0.9.14 fix added auto-`--file` for the case where the positional was an existing regular file. The other branches — missing path, directory, dead symlink — still fell through to scanning the literal positional string. Three failure shapes surfaced during the dogfood session that produced v0.9.18–v0.9.20:
+- `claudemd lint /tmp/missing-msg.txt` → exit 0, "OK no hits" (silent success masking a CI misconfiguration).
+- `claudemd lint /tmp` (existing dir) → exit 0, "OK no hits" (scans the literal string `/tmp`).
+- `claudemd lint /tmp/significantly-improved.txt` (missing) → **exit 1 with banned-vocab hit on the basename** — falsely accuses the user of writing "significantly" in a commit message they didn't even compose.
+
+Same silent-fall-through family the v0.9.14 fix targeted; the original fix was scoped only to the "file exists" branch. The third shape is the worst — a deny that points at text the user never wrote.
+
+### Fixed
+
+- `[fix]` **`bin/claudemd-lint.js#lintCmd` rejects path-shape positionals that don't resolve to a regular file.** Pre-fix: `try { stat(); if isFile read } catch { /* fall through */ }` then `text = positional.join(' ')`. Post-fix: when the positional contains `/` (or is `.` / `..`), a stat miss exits 2 with `file not found`, and a stat-as-non-file exits 2 with `is not a regular file`. Non-path-shape positionals (single word, no slash) keep the v0.9.14 text fallback — `lint significantly` MUST stay a text scan, and `lint message.txt` (no slash, missing) stays as text to preserve the pre-commit-hook ergonomic where `--file` is explicit.
+
+### Added
+
+- `[test]` **4 new cases in `tests/scripts/lint-cli.test.js`** (267 → 271 total): `lint /path/missing.txt` exits 2 with file-not-found, `lint <directory>` exits 2 with not-a-regular-file, `lint /path/with-banned-word-in-name (missing)` exits 2 (the false-positive-deny shape — anchors that the path string is NEVER scanned), `lint message.txt` (single word, no slash) stays as text scan exit 0 (anchor that the fix doesn't regress non-path-shape literals).
+
+### Why no L3 / pre-ship-review chain
+
+`fix:` per §2 hard-upgrade exclusion — restores the v0.9.14 fix's documented intent (path-shape input that doesn't resolve to a file should NOT silently scan the path string). L2 ceiling. Diff: 1 file changed in `bin/`, 1 file in `tests/scripts/`, ~25 LOC added. Pattern continuity: same antipattern family as v0.9.14/v0.9.15/v0.9.16/v0.9.17/v0.9.18 — "input parser silently falls through on unexpected shape, exits 0 looking like success." The lint-argv gate (v0.9.19) covers the JS .find/.includes/.indexOf shapes; this is a different shape (path-resolution branch missing) that lives in the lintCmd auto-detect logic, not in flag parsing — so the gate doesn't catch it. No new gate added: this branch is now covered by the four new regression tests, which is sufficient for a one-shot logic branch (the gate exists for the recurring `args.find` family, not every branch in every CLI).
+
+## [0.9.20] - 2026-05-10
+
+**Patch — `hard-rules-audit` demoteCandidates produced false-positive recommendations when log span < requested window.** Spec v6.11.7 unchanged. Surfaced in the same dogfood session that produced v0.9.18/v0.9.19: running `/claudemd-rules` (default 90-day window) against a `~/.claude/logs/claudemd.jsonl` that only spans 17 days reported `§11-memory-read` as a demote candidate — but that rule had been silently no-op'd in projects with `_` in the cwd path until v0.9.15 fixed it. The data couldn't see the rule firing because the rule itself was broken; "0 hits in window" wasn't a coldness signal. §0.1 HARD specifies "0 hits in 90d" — running the calculation on 17 days of data violates the spec. Same root applies to any rule fixed/added more recently than the log start.
+
+### Fixed
+
+- `[fix]` **`scripts/hard-rules-audit.js` suppresses `demoteCandidates` when `insufficientData`.** Pre-fix: `demoteCandidates` was computed as `hookEnforced.filter(r => r.hits.total === 0)` regardless of whether the log reached back the requested window. Post-fix: when `logSpanDays < days`, `demoteCandidates` is forced to `[]` and the would-have-been list is preserved under a new `demoteSuppressed` field (`{ reason, wouldHaveBeen }`) so the operator sees what's *potentially* cold without auto-acting on a false signal.
+
+### Added
+
+- `[feat]` **`scripts/lib/rule-hits-parse.js#logFirstTs(path)`** — returns earliest ts in the rule-hits log (ms-since-epoch) or `null` for missing/empty/all-malformed files. Skips rows with non-finite timestamps. Used by `hardRulesAudit` to compute `logSpanDays`; future consumers (sparkline, audit) can adopt for the same insufficient-data check.
+- `[feat]` **`hardRulesAudit` output: 4 new fields** — `logSpanDays` (operator transparency, surfaced even when sufficient), `insufficientData` (boolean), `demoteSuppressed` (`{reason, wouldHaveBeen}` when insufficient, `null` otherwise), and `demoteCandidates: []` when insufficient (compatibility: existing consumers reading the array get an empty array instead of false candidates).
+- `[test]` **5 new cases in `tests/scripts/hard-rules-audit.test.js`** + **8 new cases in new `tests/scripts/rule-hits-parse.test.js` (257 → 270)**: insufficient-data suppresses + surfaces `demoteSuppressed`, sufficient-span preserves existing demote behavior, log-span boundary case (35d > 30d window), `logSpanDays` always reflects actual log reach not the window, `logFirstTs` returns `null` on missing/empty/all-malformed and earliest ts otherwise, skips non-finite ts rows.
+- `[test]` **`demoteCandidates list hook-rules with zero hits` test updated** to write a 31-day-old sentinel row before asserting `§8-rm-rf-var` appears as candidate. Pre-fix, the test passed against an empty log because there was no insufficient-data check. Post-fix, an empty log triggers `insufficientData=true` and `demoteCandidates=[]`, so the test would have asserted on suppressed candidates and failed. Documented in the new "sufficient log span" test name.
+
+### Why no L3 / pre-ship-review chain
+
+`fix:` per §2 hard-upgrade exclusion — restores §0.1 HARD's documented intent (90d window means 90d of data, not "whatever the log has"). L2 ceiling. Diff: 1 lib helper added (~15 LOC), 1 script branch added (~15 LOC), 2 test files (+13 cases). Notable: this is the same family as v0.9.13's `session-summary` window-calculation bug — "report computes correct math against the data it has, but the data itself doesn't span the window the report claims to cover." Sparkline's `(newly active)` annotation is the same UX issue — not in this patch (would require updating the existing `newly active` test fixture to span 90 days, larger churn) — flagged for a follow-up patch.
+
 ## [0.9.19] - 2026-05-10
 
 **Patch — repo-wide CI guard for the argv-shape silent-fallback antipattern (§13.2 Tier 1).** Spec v6.11.7 unchanged. v0.9.14 → v0.9.18 was five consecutive patches chasing the same antipattern in five different files because each fix-set was scoped to whatever the repro session tripped on. The CHANGELOG entry for v0.9.18 stated "the next exploratory-testing session WILL find a 6th hole"; this patch makes that prediction grep-enforceable instead of trusting the next session not to slip.

package/bin/claudemd-lint.js

@@ -164,13 +164,41 @@ function lintCmd(rawArgs) {
     // from a git pre-commit hook). Without this, `claudemd lint message.txt`
     // silently scans the LITERAL STRING "message.txt" → exits 0 even when
     // the file contents would deny.
+    //
+    // v0.9.21 — close the v0.9.14 silent-fall-through residual: when the
+    // positional looks like a PATH (contains '/' or is '.' / '..') AND the
+    // path doesn't resolve to a regular file, error out instead of scanning
+    // the literal string. Pre-fix, `lint /tmp/missing.txt` scanned the path
+    // string (exit 0); `lint /tmp` (existing dir) scanned '/tmp' (exit 0);
+    // `lint /tmp/significantly-improved.txt` (missing) matched "significantly"
+    // in the basename and falsely exited 1. Same silent-success family the
+    // v0.9.14 fix targeted; the fix was scoped only to the "file exists" branch.
+    //
+    // Non-path-shape positionals (single word, no slash) keep the v0.9.14
+    // text fallback — `lint significantly` MUST stay a text scan because it's
+    // a single literal word, not a typo'd path. `lint message.txt` (no slash,
+    // missing file) stays text — it's ambiguous between "literal text" and
+    // "filename in cwd"; pre-existing-file behavior wins to preserve the
+    // pre-commit-hook ergonomic where `--file` is explicit.
     if (positional.length === 1) {
+      const arg = positional[0];
+      const looksLikePath = arg.includes('/') || arg === '.' || arg === '..';
       try {
-        const st = fs.statSync(positional[0]);
+        const st = fs.statSync(arg);
         if (st.isFile()) {
-          text = fs.readFileSync(positional[0], 'utf8');
+          text = fs.readFileSync(arg, 'utf8');
+        } else if (looksLikePath) {
+          process.stderr.write(`lint: '${arg}' is not a regular file (use --file PATH for explicit file scan or quote literal text)\n`);
+          process.exit(2);
         }
-      } catch { /* not a path — fall through to text */ }
+        // Non-path-shape + non-file (e.g. a symlink loop, fifo) → fall through to text scan.
+      } catch (e) {
+        if (looksLikePath) {
+          process.stderr.write(`lint: file not found: ${arg}\n`);
+          process.exit(2);
+        }
+        // Non-path-shape miss → fall through to text scan.
+      }
     }
     if (text === undefined) text = positional.join(' ');
   } else {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "claudemd-cli",
-  "version": "0.9.19",
+  "version": "0.9.21",
   "description": "Standalone CLI for §10-V banned-vocab + transcript scanning. Companion to the claudemd Claude Code plugin (github.com/sdsrss/claudemd) for use in git pre-commit hooks, GitHub Actions, and other agents.",
   "type": "module",
   "bin": {