Add Projects feature and comprehensive test coverage

.github/workflows/ci.yml

@@ -0,0 +1,65 @@
+name: CI
+
+on:
+  pull_request:
+    branches: [master]
+  push:
+    branches: [master]
+
+jobs:
+  lint-and-typecheck:
+    name: Lint & Typecheck
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-node@v4
+        with:
+          node-version: 22
+          cache: npm
+          cache-dependency-path: package-lock.json
+      - run: npm ci
+      - run: npm run check
+
+  unit-tests:
+    name: Unit Tests
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-node@v4
+        with:
+          node-version: 22
+          cache: npm
+          cache-dependency-path: package-lock.json
+      - run: npm ci
+      - name: Pure unit tests
+        run: |
+          npm run test:chunker
+          npm run test:vectorstore
+          npm run test:dedup
+          npm run test:smart-compile
+          npm run test:skill-sources
+          npm run test:mcp-hosts
+          npm run test:modes
+          npm run test:crypto
+          npm run test:security
+          npm run test:validation
+          npm run test:backup
+          npm run test:ranking
+          npm run test:mutex
+
+  integration-tests:
+    name: Integration Tests
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-node@v4
+        with:
+          node-version: 22
+          cache: npm
+          cache-dependency-path: package-lock.json
+      - run: npm ci
+      - name: Server + HTTP tests
+        run: |
+          npm run smoke
+          npm run test:handoffs
+          npm run test:projects

bench/README.md

@@ -10,7 +10,7 @@ For each task in `tasks.json`, three numbers (apples-to-apples on full skill bod
 - **Smart** — tokens after CE's `/api/compile/smart` picks the relevant skills for this specific task. Same content type as Raw all, just a subset.
 - **Search** — tokens an MCP host (Claude Desktop, Codex) actually pulls when it calls `context_engine_search` — chunks, not full skills.
 
-Reference column: `CONTEXT.md` size — CE's pre-compressed system-prompt summary, a *different* path entirely. Reported alongside so both CE delivery paths are visible without inflating the savings number by mixing content types.
+Reference column: `CONTEXT.md` size — CE's pre-compressed system-prompt summary, a _different_ path entirely. Reported alongside so both CE delivery paths are visible without inflating the savings number by mixing content types.
 
 Optional quality column (`--grade`): for each task, actually run it through Claude in both smart and search modes, capture the response, and have a judge model score it 1-10. The summary adds tokens-per-quality-point efficiency — the real "are we saving tokens AND maintaining answer quality?" question.
 
@@ -64,7 +64,7 @@ Cost expectation at defaults (Haiku for both task + grader, 15 tasks): ~$0.10 pe
 
 - **Tokenizer**: tiktoken `cl100k_base` (the BPE GPT-3.5/4 uses). It's within ~5% of Anthropic's own tokenizer on prose. Same encoding is used for baseline + smart + search counts, so the percentages are internally consistent.
 - **Baseline definition**: only **active** skills (per `data/skill-states.json`) — that's the realistic ceiling. Including inactive skills would inflate the saving artificially.
-- **Search realism**: simulates a host calling `context_engine_search` *once* per task and pulling N chunks. Real host apps may call it multiple times or fall back to `context_engine_get_skill` for full bodies, so this is a lower-bound on what they actually consume.
+- **Search realism**: simulates a host calling `context_engine_search` _once_ per task and pulling N chunks. Real host apps may call it multiple times or fall back to `context_engine_get_skill` for full bodies, so this is a lower-bound on what they actually consume.
 - **Smart-compile token budget**: defaults to 16k. Run with `--max-tokens 4000` to see how aggressive selection becomes for small-context models, or `--max-tokens 64000` to see what large-context users get.
 - **Cross-check**: the summary prints CE's own estimator's ratio vs tiktoken's count. If those diverge a lot, CE's internal budget UI is over- or under-stating.
 

cli/index.js

@@ -434,4 +434,4 @@ function help() {
     context-engine add ./my-skills/react-rules
     context-engine status
   `);
-}
+}

data/projects.json

@@ -0,0 +1,4 @@
+{
+  "version": "1.0",
+  "projects": []
+}

docs/specs/onboarding-redesign.md

@@ -28,12 +28,12 @@ The sibling product `model-db` ships a polished onboarding wizard. Its pattern 
 
 Four steps, each focused on one decision.
 
-| # | Step       | Concern                                                                       | Replaces                                |
-| - | ---------- | ----------------------------------------------------------------------------- | --------------------------------------- |
-| 1 | **Connect**| Pick which MCP hosts (Claude Desktop, Codex CLI, ChatGPT) to wire CE into.    | Most of current "Discover" + "Connect". |
-| 2 | **Context**| Review skills/memory/index. Optional inline "Build index" if Ollama present.  | "Available context" panel + part of "Health". |
-| 3 | **IDE**    | Show detected fallback targets. Note that file output remains available.      | "IDE and file-output surfaces" panel.   |
-| 4 | **Health** | Quick verification: hosts connected, skills active, index ready. Finish.      | Current "Health" step.                  |
+| #   | Step        | Concern                                                                      | Replaces                                      |
+| --- | ----------- | ---------------------------------------------------------------------------- | --------------------------------------------- |
+| 1   | **Connect** | Pick which MCP hosts (Claude Desktop, Codex CLI, ChatGPT) to wire CE into.   | Most of current "Discover" + "Connect".       |
+| 2   | **Context** | Review skills/memory/index. Optional inline "Build index" if Ollama present. | "Available context" panel + part of "Health". |
+| 3   | **IDE**     | Show detected fallback targets. Note that file output remains available.     | "IDE and file-output surfaces" panel.         |
+| 4   | **Health**  | Quick verification: hosts connected, skills active, index ready. Finish.     | Current "Health" step.                        |
 
 Each step has Back / Next (or Skip for now / Finish on the last step). Steps 1 and 3 are skippable — the user can connect later via the Connections tab.
 
@@ -67,11 +67,11 @@ No `.onboarding-mark` text monogram. No typeset "Context Engine" string paired w
 
 Horizontal row of numbered circles connected by short lines. State per step:
 
-| State    | Circle background    | Circle text     | Connector line  |
-| -------- | -------------------- | --------------- | --------------- |
-| Done     | `--accent`           | white           | `--accent`      |
-| Current  | `--accent`           | white           | `--dram-line-subtle` (line *after* current) |
-| Upcoming | `--dram-bg-recessed` | `--text-4`      | `--dram-line-subtle` |
+| State    | Circle background    | Circle text | Connector line                              |
+| -------- | -------------------- | ----------- | ------------------------------------------- |
+| Done     | `--accent`           | white       | `--accent`                                  |
+| Current  | `--accent`           | white       | `--dram-line-subtle` (line _after_ current) |
+| Upcoming | `--dram-bg-recessed` | `--text-4`  | `--dram-line-subtle`                        |
 
 Connector: `height: 1px; width: 32px;`. Circle: `28×28; border-radius: 999px`. Step label sits under each circle in `--text-3`, `11px`, uppercase, `letter-spacing: 0.07em` (matches card-name typography from DRAM).
 
@@ -97,16 +97,16 @@ Common pattern inside each step:
 
 All cards in the onboarding body use the global card pattern from [dram-design-map.md → Cards](../dram-design-map.md#cards):
 
-| Property           | Token                       |
-| ------------------ | --------------------------- |
-| Background         | `--dram-card-bg`            |
-| Border             | `--dram-card-border`        |
-| Hover background   | `--dram-card-hover`         |
-| Hover border       | `--dram-card-border-hover`  |
-| Selected background| `--dram-card-active`        |
-| Selected border    | `--dram-card-border-active` |
-| Radius             | `--r-3`                     |
-| Disabled opacity   | `--dram-card-muted-opacity` |
+| Property            | Token                       |
+| ------------------- | --------------------------- |
+| Background          | `--dram-card-bg`            |
+| Border              | `--dram-card-border`        |
+| Hover background    | `--dram-card-hover`         |
+| Hover border        | `--dram-card-border-hover`  |
+| Selected background | `--dram-card-active`        |
+| Selected border     | `--dram-card-border-active` |
+| Radius              | `--r-3`                     |
+| Disabled opacity    | `--dram-card-muted-opacity` |
 
 Three card variants used:
 

docs/specs/skill-sources.md

@@ -61,13 +61,13 @@ A registered external directory CE reads `SKILL.md` files from. Stored in `data/
 - `removeSource(id)` → deletes by id. Refuses if id is `internal`.
 - `scanHostSkillPaths()` → probes known locations, returns `{ path, label, exists, skillCount }[]`. Probed paths:
 
-  | Path                                  | Label                       |
-  | ------------------------------------- | --------------------------- |
-  | `~/.claude/skills`                    | Claude Code (global)        |
-  | `<CWD>/.claude/skills`                | Claude Code (current project)|
-  | `<CWD>/.clinerules`                   | Cline / Roo rules           |
-  | `<CWD>/.continue/rules`               | Continue.dev rules          |
-  | `~/.opencode/skills`                  | OpenCode (global)           |
+  | Path                    | Label                         |
+  | ----------------------- | ----------------------------- |
+  | `~/.claude/skills`      | Claude Code (global)          |
+  | `<CWD>/.claude/skills`  | Claude Code (current project) |
+  | `<CWD>/.clinerules`     | Cline / Roo rules             |
+  | `<CWD>/.continue/rules` | Continue.dev rules            |
+  | `~/.opencode/skills`    | OpenCode (global)             |
 
   Each entry returns the resolved absolute path, a friendly label, `exists` boolean, and skill count if any.
 
@@ -99,13 +99,13 @@ The denylist is reused but **inverted** for this case: we're reading, not writin
 
 ### New endpoints
 
-| Method | Path                          | Description                                                                    |
-| ------ | ----------------------------- | ------------------------------------------------------------------------------ |
-| GET    | `/api/skill-sources`          | List registered sources + implicit internal. Returns skill counts per source.  |
-| POST   | `/api/skill-sources`          | Add an external source. Body: `{ path, label? }`. Validates + denylist-checks. |
-| DELETE | `/api/skill-sources/:id`      | Remove a source. Refuses on `internal`.                                        |
-| GET    | `/api/skill-sources/scan`     | Probe known host-skill paths. Returns the candidates with counts.              |
-| POST   | `/api/skill-sources/import`   | Phase 2 — copy a source's contents into `<CE_ROOT>/skills/imported/<id>/`.     |
+| Method | Path                        | Description                                                                    |
+| ------ | --------------------------- | ------------------------------------------------------------------------------ |
+| GET    | `/api/skill-sources`        | List registered sources + implicit internal. Returns skill counts per source.  |
+| POST   | `/api/skill-sources`        | Add an external source. Body: `{ path, label? }`. Validates + denylist-checks. |
+| DELETE | `/api/skill-sources/:id`    | Remove a source. Refuses on `internal`.                                        |
+| GET    | `/api/skill-sources/scan`   | Probe known host-skill paths. Returns the candidates with counts.              |
+| POST   | `/api/skill-sources/import` | Phase 2 — copy a source's contents into `<CE_ROOT>/skills/imported/<id>/`.     |
 
 ## UI
 
@@ -167,6 +167,7 @@ A new "Sources" affordance in the Skills tab header — a small select/expander
 ## Phase 2 detailed design (2026-05-11)
 
 Locked decisions:
+
 - **Import action lives on the linked row.** Link is the cheap commitment, Import is the heavier one; forcing Link-then-Import is the right progression. No Import button on candidate rows.
 - **Manual Sync only.** No filesystem watching; a Sync button on imported rows triggers the diff. Keeps the user in control of when CE walks their dirs.
 
@@ -175,6 +176,7 @@ Locked decisions:
 Three states for a registered source: `external` (linked, read from original), `imported` (linked + a copy/hard-link tree has been written into `<CE_ROOT>/skills/imported/<sourceId>/`), `internal` (implicit, never stored).
 
 Transitions:
+
 ```
 external --[POST /api/skill-sources/:id/import]--> imported
 imported --[POST /api/skill-sources/:id/sync/apply]--> imported (manifest rewritten)
@@ -187,6 +189,7 @@ Unlinking an imported source intentionally **keeps** the imported tree. The user
 ### Import strategy: hard-link + copy fallback
 
 File-level `fs.linkSync` is the primary strategy. Fall back to `fs.copyFileSync` per-file on:
+
 - `EXDEV` — cross-volume on Windows. NTFS hard-links are intra-volume only.
 - `EPERM` / `EACCES` — source permissions.
 - Non-link-capable filesystems (FAT/exFAT).
@@ -205,9 +208,7 @@ One file per imported source at `data/skill-imports/<sourceId>.json`:
   "importedAt": "2026-05-11T15:00:00Z",
   "lastSyncedAt": "2026-05-11T15:00:00Z",
   "aggregateStrategy": "link",
-  "files": [
-    { "rel": "react/SKILL.md", "size": 2341, "mtimeMs": 1715000000000, "strategy": "link" }
-  ]
+  "files": [{ "rel": "react/SKILL.md", "size": 2341, "mtimeMs": 1715000000000, "strategy": "link" }]
 }
 ```
 
@@ -226,23 +227,25 @@ One file per imported source at `data/skill-imports/<sourceId>.json`:
 ```
 
 Definitions:
+
 - **Added**: in source, absent from manifest.
 - **Removed**: in manifest, absent from source.
 - **Modified**: in both, but `size` or `mtimeMs` differs **and** the per-file strategy is `copy`. Hard-linked files can't drift in content (shared inode), so we don't list them as modified even if mtime moved on the source.
 
 `POST /api/skill-sources/:id/sync/apply` body `{ mode: 'append' | 'overwrite' }`:
+
 - `append`: adds new files only. Removed and modified are left.
 - `overwrite`: applies all three categories — add new, delete removed, re-link/re-copy modified.
 
 Manifest is rewritten after successful apply.
 
 ### Endpoints (Phase 2)
 
-| Method | Path                                            | Description                                                                |
-| ------ | ----------------------------------------------- | -------------------------------------------------------------------------- |
-| POST   | `/api/skill-sources/:id/import`                 | Walk source + write imported tree + manifest. Idempotent: refuses if already imported. |
-| GET    | `/api/skill-sources/:id/sync`                   | Compute diff without applying.                                             |
-| POST   | `/api/skill-sources/:id/sync/apply`             | Body `{ mode }`. Apply diff + rewrite manifest.                            |
+| Method | Path                                | Description                                                                            |
+| ------ | ----------------------------------- | -------------------------------------------------------------------------------------- |
+| POST   | `/api/skill-sources/:id/import`     | Walk source + write imported tree + manifest. Idempotent: refuses if already imported. |
+| GET    | `/api/skill-sources/:id/sync`       | Compute diff without applying.                                                         |
+| POST   | `/api/skill-sources/:id/sync/apply` | Body `{ mode }`. Apply diff + rewrite manifest.                                        |
 
 ### UI placement (Phase 2)
 
@@ -306,6 +309,7 @@ Verification (direct node script): linking a fixture with both an `app-launcher`
 - `conflicts`: dest AND source both diverged. Overwrite discards the local edit.
 
 Apply behaviour:
+
 - `append` ignores both.
 - `overwrite` re-places `modified` + `localEdits` + `conflicts` (excluding conflicts whose source file no longer exists; those flow through `removed`).
 

electron/main.cjs

@@ -345,9 +345,9 @@ ipcMain.on('window:close', () => {
 ipcMain.handle('dialog:select-folder', async (_event, options) => {
   const dialogOptions = {
     properties: ['openDirectory'],
-    title: (options && typeof options.title === 'string' ? options.title : 'Pick a folder'),
+    title: options && typeof options.title === 'string' ? options.title : 'Pick a folder',
   };
   const result = await dialog.showOpenDialog(mainWindow || undefined, dialogOptions);
   if (result.canceled || !result.filePaths?.length) return null;
   return result.filePaths[0];
-});
+});

electron/preload.cjs

@@ -31,4 +31,4 @@ contextBridge.exposeInMainWorld('contextEngineDesktop', {
 
 window.addEventListener('DOMContentLoaded', () => {
   document.documentElement.dataset.runtime = 'electron';
-});
+});

electron/updater.cjs

@@ -88,4 +88,4 @@ function startAutoUpdate(window, options = {}) {
   }, intervalMs);
 }
 
-module.exports = { startAutoUpdate };
+module.exports = { startAutoUpdate };

mcp-http-server.mjs

@@ -175,4 +175,4 @@ server.listen(PORT, HOST, () => {
     `context-engine HTTP MCP server listening on http://${HOST}:${PORT}/mcp (auth=${auth}, CE_BASE=${CE_BASE})\n`,
   );
   if (oauth) process.stderr.write('OAuth consent passphrase is set with MCP_OAUTH_PASSWORD.\n');
-});
+});

mcp-oauth.mjs

@@ -299,4 +299,4 @@ async function collect(req) {
   const chunks = [];
   for await (const chunk of req) chunks.push(chunk);
   return chunks;
-}
+}

mcp-server.mjs

@@ -14,4 +14,4 @@ const transport = new StdioServerTransport();
 await server.connect(transport);
 
 // stderr is the only safe channel under stdio MCP; stdout is protocol traffic.
-process.stderr.write(`context-engine MCP server connected (CE_BASE=${CE_BASE})\n`);
+process.stderr.write(`context-engine MCP server connected (CE_BASE=${CE_BASE})\n`);

package.json

@@ -1,6 +1,6 @@
 {
   "name": "context-engine",
-  "version": "0.3.1",
+  "version": "0.4.0",
   "description": "Local-first continuity layer for AI work across tools, providers, and fresh sessions.",
   "main": "electron/main.cjs",
   "bin": {
@@ -40,6 +40,14 @@
     "test:skill-sources": "node scripts/skill-sources-smoke.js",
     "migrate:handoffs": "node scripts/migrate-legacy-handoff.js",
     "test:mcp-hosts": "node scripts/mcp-host-config-smoke.js",
+    "test:modes": "node scripts/mode-apply-smoke.js",
+    "test:projects": "node scripts/projects-smoke.js",
+    "test:crypto": "node scripts/crypto-smoke.js",
+    "test:security": "node scripts/security-smoke.js",
+    "test:validation": "node scripts/validation-smoke.js",
+    "test:backup": "node scripts/backup-smoke.js",
+    "test:ranking": "node scripts/ranking-smoke.js",
+    "test:mutex": "node scripts/mutex-smoke.js",
     "mcp": "node mcp-server.mjs",
     "mcp:http": "node mcp-http-server.mjs",
     "mcpb:pack": "powershell -NoProfile -ExecutionPolicy Bypass -File scripts/pack-mcpb.ps1",