Release D: GIT-113 Phase 1 (GIT-114 to GIT-120)

.github/workflows/ci.yml

@@ -51,6 +51,17 @@ jobs:
       - name: E2E — session identity survives MCP restart
         run: npx vitest run --config vitest.e2e.config.ts tests/e2e/git-89-session-identity.test.ts
 
+      # GIT-120: install-hooks / uninstall-hooks keep foreign hooks. Runs the
+      # real CLI in a scratch repo; no server, Docker or credentials.
+      - name: E2E — hooks install/uninstall keep foreign hooks
+        run: npx vitest run --config vitest.e2e.config.ts tests/e2e/git-120-hooks-merge.test.ts
+
+      # GIT-115: a fresh install in a clean HOME — init, the SessionStart hook,
+      # the built server — plus the CLI/wizard suite. Both run the real CLI in
+      # scratch directories with a scratch HOME; no Docker or credentials.
+      - name: E2E — fresh install reads what init wrote; CLI and wizard
+        run: npx vitest run --config vitest.e2e.config.ts tests/e2e/git-115-fresh-install.test.ts tests/e2e/cli-fresh-install.test.ts
+
       - name: Dependency audit (no unused deps)
         run: npx depcheck --ignores="@types/*" --json | node -e "const d=JSON.parse(require('fs').readFileSync('/dev/stdin','utf8')); const unused=d.dependencies||[]; if(unused.length){console.error('Unused dependencies:',unused);process.exit(1)}"
 
@@ -73,6 +84,31 @@ jobs:
       - name: Hook tests
         run: bash hooks/tests/test-hooks.sh
 
+  # GIT-116: the same suite on stock macOS. The runner has Homebrew (and may
+  # have coreutils' timeout), so the suite runs with the system directories
+  # plus a directory holding only node — what a Mac without Homebrew has.
+  hooks-macos:
+    runs-on: macos-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: actions/setup-node@v4
+        with:
+          node-version: 20
+
+      - name: Hook tests on a stock macOS PATH (no Homebrew, no timeout)
+        run: |
+          mkdir -p "$RUNNER_TEMP/nodeonly"
+          ln -s "$(command -v node)" "$RUNNER_TEMP/nodeonly/node"
+          STOCK="/usr/bin:/bin:/usr/sbin:/sbin:$RUNNER_TEMP/nodeonly"
+          for t in timeout gtimeout; do
+            if PATH="$STOCK" command -v "$t" >/dev/null; then echo "::error::$t is on the stock PATH"; exit 1; fi
+          done
+          # macOS 15+ ships /usr/bin/jq; older macOS does not. The jq-less path
+          # is covered by hooks-without-jq (Linux); report which this runner is.
+          PATH="$STOCK" command -v jq >/dev/null && echo "stock jq present: $(PATH="$STOCK" jq --version)" || echo "stock jq absent"
+          env -i HOME="$HOME" PATH="$STOCK" /bin/bash hooks/tests/test-hooks.sh
+
   # GIT-83: the release gate. Customers never run SQL after setup (standing
   # rule, 2026-09-21), so a release must work on a store still on the v1.8.0
   # setup.sql — the venue is put on exactly that schema first, then the
@@ -164,7 +200,7 @@ jobs:
           retention-days: 14
 
   publish:
-    needs: [build, hooks-without-jq, venue-gate]
+    needs: [build, hooks-without-jq, hooks-macos, venue-gate]
     runs-on: ubuntu-latest
     if: startsWith(github.ref, 'refs/tags/v')
     permissions:

CHANGELOG.md

@@ -7,6 +7,54 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [1.11.1] - 2026-09-22
+
+**Just upgrade the package. Nothing to change on your database.** Every fix below works on a project
+still on the `setup.sql` you first ran, back to 1.8.0, and was checked against a project on exactly
+that schema and on the current one.
+
+**Recall works when it has to ask your database.** When the local index was not ready, or with
+`GITMEM_SEARCH_MODE=remote`, recall called a database function that `setup.sql` never created. It
+got an error and returned no scars. It now calls `gitmem_scar_search`, which every project has, and
+falls back to the older function only where that exists. (GIT-114)
+
+**A fresh install now finds its starter lessons.** `gitmem init` put your memory in the repo's
+`.gitmem/`, but gitmem reads `~/.gitmem`, so a new install started with none of its starter lessons
+and was told its memory was "NOT being read". `init` now puts your memory where gitmem reads it
+(`GITMEM_DIR`, else `GITMEM_HOME/.gitmem`, else `~/.gitmem`). The repo's `.gitmem/` keeps only
+`config.json`, with the project name, and the hook scripts. A repo set up that way no longer
+triggers the warning. `gitmem uninstall` removes the repo's files and keeps your memory, which every
+project shares, unless you pass `--all`. The session-start hook now finds the project name without
+jq. (GIT-115)
+
+If an older `init` put lessons of your own in a repo's `.gitmem/`, session_start still points them
+out. `npx gitmem-mcp migrate-root` copies them into `~/.gitmem`.
+
+**`install-hooks` and `uninstall-hooks` keep your own hooks.** They used to replace or delete every
+hook in `.claude/settings.json` or `.cursor/hooks.json`. Now they add or remove only gitmem's hooks,
+back the file up first, and print what changed. A file that is not valid JSON is left alone instead
+of being replaced. `gitmem init` and `gitmem uninstall` use the same matching, so a hook of yours
+that shares a matcher with gitmem's, or lives in a folder named gitmem, is no longer removed or
+mistaken for gitmem's. (GIT-120, GIT-115)
+
+### Fixed
+
+- **Hooks work on a stock Mac.** The prompt hook ran `timeout`, which macOS does not have, so it
+  never returned scars there. The time limit now lives inside gitmem. When gitmem or node cannot be
+  found, the hook says so instead of staying silent. (GIT-116)
+- **A thread that failed to sync is no longer marked synced.** `session_close` reports PARTIAL and
+  names the thread ids that did not reach your database. Those threads stay in `threads.json` and
+  the next session tries again. (GIT-117)
+- **A failed index reload no longer empties recall.** If reloading scars from your database fails,
+  gitmem keeps the index it had, falls back to the on-disk cache, and retries in the background.
+  `create_learning` says when the new learning is saved but not yet searchable. (GIT-118)
+- **`archive_learning` accepts a short id.** Archiving by id prefix failed on every project with
+  "operator does not exist". Archiving an id that does not exist used to report success; it now says
+  nothing was archived. (GIT-119)
+- **Updates that change no row are reported as not saved.** This covers thread resolve, touch and
+  auto-archive, session embeddings, superseded sessions, relevance data and transcript paths. Each
+  one used to count as success. (GIT-119)
+
 ## [1.11.0] - 2026-09-21
 
 **Just upgrade the package. Nothing to change on your database.** Every fix in this release works on

bin/gitmem-root.js

@@ -0,0 +1,34 @@
+/**
+ * GIT-115: where the CLI puts things, split the same way the server reads them.
+ *
+ * The store (learnings, threads, sessions, decisions, the closing payload and
+ * its template) belongs in the root the server reads. That is resolved exactly
+ * as src/services/gitmem-dir.ts does, without a running session: GITMEM_DIR,
+ * then GITMEM_HOME/.gitmem, then ~/.gitmem. `init` used to seed <cwd>/.gitmem,
+ * which the server has not read since GIT-91, so a fresh install saw no
+ * starter lessons and was told its store was "NOT being read".
+ *
+ * The repo keeps its own .gitmem/ for two things only: config.json with the
+ * project name (the SessionStart hook reads it from the repo) and the hook
+ * scripts (committed settings refer to them by relative path).
+ */
+
+import { join } from "path";
+import { homedir } from "os";
+
+/** The root the server reads. */
+export function storeRoot() {
+  if (process.env.GITMEM_DIR) return process.env.GITMEM_DIR;
+  return join(process.env.GITMEM_HOME || homedir(), ".gitmem");
+}
+
+/** The repo's .gitmem/: config.json and hooks/ only. */
+export function repoGitmemDir(cwd = process.cwd()) {
+  return join(cwd, ".gitmem");
+}
+
+/** A path shortened to ~ for display. */
+export function displayPath(p) {
+  const home = homedir();
+  return p === home || p.startsWith(home + "/") ? "~" + p.slice(home.length) : p;
+}