feat: add e2e pipeline with publint, attw, and cleanup

Add package structure validation (publint --strict) and TypeScript type
resolution checks (attw) to the e2e pipeline. Artifacts (.tgz,
e2e/node_modules, e2e/package-lock.json) are now cleaned up after every
run. E2e job added to CI in parallel with existing jobs, gating publish.

Author: SimplyLiz

.github/workflows/ci.yml

@@ -52,8 +52,31 @@ jobs:
           fi
       - run: npx tsc --noEmit
 
+  bench:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v6
+      - uses: actions/setup-node@v6
+        with:
+          node-version: 22
+          cache: npm
+      - run: npm ci
+      - run: npm run bench:check
+
+  e2e:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v6
+      - uses: actions/setup-node@v6
+        with:
+          node-version: 22
+          cache: npm
+      - run: npm ci
+      - name: Build, pack, lint, and smoke test
+        run: npm run test:e2e
+
   publish:
-    needs: [audit, lint, test]
+    needs: [audit, lint, test, e2e]
     if: startsWith(github.ref, 'refs/tags/v')
     runs-on: ubuntu-latest
     permissions:

.gitignore

@@ -5,9 +5,11 @@ venv/
 
 # build
 dist/
+coverage/
 build/
 *.egg-info/
 *.tsbuildinfo
+*.tgz
 
 # cache
 __pycache__/

e2e/.gitignore

@@ -0,0 +1,2 @@
+node_modules/
+package-lock.json

e2e/README.md

@@ -0,0 +1,62 @@
+# E2E Smoke Test
+
+Installs `context-compression-engine` as a real consumer would and exercises every public export.
+
+Catches issues that unit tests can't: broken `exports` map, missing files in the tarball, ESM resolution failures, async path regressions.
+
+## Pipeline
+
+```
+npm run test:e2e
+```
+
+Runs: **build → pack → publint + attw → smoke test → cleanup**
+
+| Step | What it does |
+|------|-------------|
+| `npm run build` | Compile TypeScript |
+| `npm pack` | Create tarball from `files` field |
+| `publint --strict` | Validate package.json exports, files, types |
+| `attw` | Check TypeScript type resolution across all `moduleResolution` settings |
+| `smoke.mjs` | 68 assertions exercising the public API |
+| cleanup | Remove `.tgz`, `e2e/node_modules`, `e2e/package-lock.json` |
+
+Cleanup always runs, even on failure. The exit code from the smoke test is preserved.
+
+## Other scripts
+
+```bash
+# Test the published npm package (post-publish validation)
+npm run test:e2e:published
+```
+
+## What the smoke test covers
+
+| # | Area | What's tested |
+|---|------|---------------|
+| 1 | Basic compress | ratio, token_ratio, message count, verbatim store |
+| 2 | Uncompress round-trip | lossless content restoration |
+| 3 | Dedup | exact duplicate detection (>=200 char messages) |
+| 4 | Token budget (fit) | binary search finds a recencyWindow that fits |
+| 5 | Token budget (tight) | correctly reports `fits: false` when impossible |
+| 6 | defaultTokenCounter | returns positive number |
+| 7 | Preserve keywords | keywords retained in compressed output |
+| 8 | sourceVersion | flows into compression metadata |
+| 9 | embedSummaryId | summary_id embedded in compressed content |
+| 10 | Factory functions | createSummarizer, createEscalatingSummarizer exported |
+| 11 | forceConverge | best-effort truncation, no regression |
+| 12 | Fuzzy dedup | runs without errors, message count preserved |
+| 13 | Provenance metadata | _cce_original structure (ids, summary_id, version) |
+| 14 | Missing verbatim store | missing_ids reported correctly |
+| 15 | Custom tokenCounter | invoked and used for ratio calculation |
+| 16 | Edge cases | empty input, single message |
+| 17 | Async path (mock summarizer) | compress returns Promise, summarizer called, round-trip works |
+| 18 | Async + token budget | async binary search produces fits/tokenCount/recencyWindow |
+| 19 | System role | system messages auto-preserved, never compressed |
+| 20 | tool_calls | messages with tool_calls pass through intact |
+| 21 | Re-compression | compress already-compressed output, recover via chained stores |
+| 22 | Recursive uncompress | nested provenance fully expanded |
+| 23 | minRecencyWindow | floor enforced during budget binary search |
+| 24 | Large conversation (31 msgs) | compression + lossless round-trip at scale |
+| 25 | Large conversation + budget | binary search converges on 50% budget target |
+| 26 | Verbatim store as object | uncompress accepts plain Record, not just function |

e2e/package.json

@@ -0,0 +1,9 @@
+{
+  "name": "cce-e2e",
+  "private": true,
+  "type": "module",
+  "description": "End-to-end smoke test — installs context-compression-engine from npm and exercises the public API as a real consumer would.",
+  "scripts": {
+    "test": "node smoke.mjs"
+  }
+}