docs: add changelog-accuracy lesson and pre-tag eval

CLAUDE.md

@@ -356,6 +356,7 @@ Keep rules general, evidence-backed, and under 20 words. Never add secrets, pers
 - Never `fetch` a provider from content/popup — route through the background worker.
 - Bump `version` in `package.json` only (+ lockfile); the manifest derives it via `pkg.version`.
 - Style inline; Tailwind classes do nothing until PostCSS is wired.
+- Verify CHANGELOG entries against code, not commit subjects or handoff summaries.
 
 ## Memory protocol
 

docs/EVALS.md

@@ -30,7 +30,14 @@
 
 ## Active failure-derived checks
 
-_No failure-derived checks yet. Add one here when a verified regression gives a deterministic trigger — e.g. a guard that fails if the built `content.js` still contains `[LexAI` logs (T-03), or a test asserting the `onMessage` listener returns `true`._
+### E-RELEASE-01 — CHANGELOG version match and spot-check
+
+- **Prevents:** L-RELEASE-01 — false release notes shipped to CWS
+- **How to run:** (1) Extract version from `package.json` (e.g., `jq -r .version package.json`). (2) Grep for `## [version]` in `CHANGELOG.md`. (3) Pick 2–3 user-visible claims (feature name, behavior, action added) and verify against `git log --oneline` or the code (`src/lib/actions.ts`, `entrypoints/*/`).
+- **Pass condition:** (1) CHANGELOG has a section header matching the version; (2) each spot-checked claim is present in code or the latest commit subject(s) describe that feature being added.
+- **Cost:** fast (~2 min).
+- **When to run:** before `git tag v*.*.*`.
+- **Last verified:** 2026-08-12 (caught two false claims in v1.1.0 prep).
 
 ## Eval template
 

docs/LESSONS_LEARNED.md

@@ -53,6 +53,12 @@
 - **Root cause / failure boundary:** Tailwind is installed but WXT PostCSS was never wired; Tailwind classes silently do nothing.
 - **Prevention:** style with inline objects and the existing dark palette; don't add Tailwind classes unless the task is explicitly to wire PostCSS.
 
+### L-RELEASE-01 — Verify CHANGELOG against code before tagging
+
+- **Root cause / failure boundary:** release notes authored from commit subjects and handoff summaries are not facts; two false claims in v1.1.0 CHANGELOG were caught pre-tag: (1) off-by-one count of actions (`12 … plus Auto` when auto is one of 12), (2) feature listed for 1.0.1 that doesn't exist in code (commit subject claimed it but Options.tsx has no such tab).
+- **Prevention:** before `git tag v*.*.*`, verify at least 2–3 user-visible changes claimed in CHANGELOG against the actual code diff or feature. Check version in CHANGELOG matches `package.json`.
+- **Eval:** E-RELEASE-01.
+
 ## Recording policy
 
 Add a lesson only after a material, evidenced learning signal (correction, unexpected failure, regression, rejected review, proven wrong assumption). Each needs a durable prevention; link a deterministic eval when possible. No secrets, credentials, personal data, or raw transcripts.