fix(continual-learning): write memory to AGENTS.local.md in team repos

.cursor-plugin/marketplace.json

@@ -16,7 +16,7 @@
     {
       "name": "continual-learning",
       "source": "continual-learning",
-      "description": "Incremental transcript-driven memory updates for AGENTS.md using high-signal bullet points only."
+      "description": "Incrementally learns durable user preferences and workspace facts from transcript changes and writes them to user-scoped AGENTS.local.md (safe for team repos)."
     },
     {
       "name": "cursor-team-kit",

continual-learning/.cursor-plugin/plugin.json

@@ -1,8 +1,8 @@
 {
   "name": "continual-learning",
   "displayName": "Continual Learning",
-  "version": "1.0.0",
-  "description": "Incrementally learns durable user preferences and workspace facts from transcript changes and keeps AGENTS.md up to date with plain bullet points.",
+  "version": "1.1.0",
+  "description": "Incrementally learns durable user preferences and workspace facts from transcript changes and writes them to user-scoped AGENTS.local.md (safe for team repos).",
   "author": {
     "name": "Eric Zakariasson"
   },

continual-learning/CHANGELOG.md

@@ -0,0 +1,13 @@
+# Changelog
+
+## 1.1.0
+
+- Default memory file is `~/.cursor/projects/<slug>/AGENTS.local.md`.
+- Stop-hook followup embeds resolved absolute paths and forbids writing the
+  repo-tracked `AGENTS.md` unless an allowed shared workspace file is set.
+- `preToolUse` denies Write/StrReplace of `## Learned *` sections into
+  repo-tracked `AGENTS.md` / `CLAUDE.md` / `GEMINI.md`, except the configured
+  workspace file when it is gitignored or `CONTINUAL_LEARNING_ALLOW_SHARED=1`.
+- Cadence/index default to `~/.cursor/projects/<slug>/continual-learning/`.
+  Legacy files are copied; tracked originals are not deleted.
+- Keep `## Learned Workspace Facts` (no `(local)` heading rename).

continual-learning/README.md

@@ -1,16 +1,33 @@
 # Continual Learning
 
-Automatically and incrementally keeps `AGENTS.md` up to date from transcript changes.
+Automatically and incrementally keeps a **user-scoped** memory file up to date
+from transcript changes. Safe to enable in repositories that already track a
+team-owned `AGENTS.md`.
+
+Default write target:
+
+```text
+~/.cursor/projects/<slug>/AGENTS.local.md
+```
+
+`<slug>` is the absolute workspace path with the leading `/` stripped and `/`
+replaced by `-` (same layout as `agent-transcripts/`). Cursor stacks this file
+with the repo `AGENTS.md`.
 
 The plugin combines:
 
-- A `stop` hook that decides when to trigger learning.
+- A `stop` hook that decides when to trigger learning and embeds absolute
+  memory/index paths in the followup.
 - A `continual-learning` skill that orchestrates the learning flow.
-- An `agents-memory-updater` subagent that mines new or changed transcripts and updates `AGENTS.md`.
+- An `agents-memory-updater` subagent that mines new or changed transcripts
+  and updates `AGENTS.local.md`.
+- A `preToolUse` hook that denies Write/StrReplace of `## Learned *` sections
+  into a repo-tracked `AGENTS.md` / `CLAUDE.md` / `GEMINI.md`, unless that path
+  is an explicit shared workspace file that is allowed.
 
 It is designed to avoid noisy rewrites by:
 
-- Reading existing `AGENTS.md` first and updating matching bullets in place.
+- Reading the existing memory file first and updating matching bullets in place.
 - Processing only new or changed transcript files.
 - Writing plain bullet points only (no evidence/confidence metadata).
 
@@ -20,19 +37,43 @@ It is designed to avoid noisy rewrites by:
 /add-plugin continual-learning
 ```
 
+## Safe defaults for team repos
+
+Learned bullets no longer go into the workspace `AGENTS.md` by default. That
+file is often owned by the whole team; writing personal preferences there
+creates dirty trees and accidental commits.
+
+Cadence and the incremental transcript index also default to:
+
+```text
+~/.cursor/projects/<slug>/continual-learning/
+```
+
+instead of `.cursor/hooks/state/` inside the repo.
+
+Section titles stay:
+
+- `## Learned User Preferences`
+- `## Learned Workspace Facts`
+
+Existing `AGENTS.local.md` files keep working; headings are not renamed.
+
 ## How it works
 
-On eligible `stop` events, the hook may emit a `followup_message` that asks the agent to run the `continual-learning` skill.
+On eligible `stop` events, the hook may emit a `followup_message` that asks the
+agent to run the `continual-learning` skill.
 
-The skill is marked `disable-model-invocation: true`, so it will not be auto-selected during normal model invocation. When it does run, it delegates the full memory update flow to `agents-memory-updater`.
+The skill is marked `disable-model-invocation: true`, so it will not be
+auto-selected during normal model invocation. When it does run, it delegates
+the full memory update flow to `agents-memory-updater`.
 
 The hook keeps local runtime state in:
 
-- `.cursor/hooks/state/continual-learning.json` (cadence state)
+- `~/.cursor/projects/<slug>/continual-learning/cadence.json`
 
 The updater uses an incremental transcript index at:
 
-- `.cursor/hooks/state/continual-learning-index.json`
+- `~/.cursor/projects/<slug>/continual-learning/index.json`
 
 ## Trigger cadence
 
@@ -48,7 +89,22 @@ Trial mode defaults (enabled in this plugin hook config):
 - minimum 15 minutes
 - automatically expires after 24 hours, then falls back to default cadence
 
-## Optional env overrides
+## Configuration (all optional)
+
+**Memory targets**
+
+- `CONTINUAL_LEARNING_USER_FILE` — personal memory file. Default:
+  `~/.cursor/projects/<slug>/AGENTS.local.md`.
+- `CONTINUAL_LEARNING_WORKSPACE_FILE` — opt-in shared file inside the repo.
+  Unset by default. Refused when the path is in git and not gitignored,
+  unless `CONTINUAL_LEARNING_ALLOW_SHARED=1`.
+- `CONTINUAL_LEARNING_STATE_DIR` — `cadence.json` + `index.json`. Default:
+  `~/.cursor/projects/<slug>/continual-learning/`.
+- `CONTINUAL_LEARNING_ALLOW_SHARED` — bypass the git-leak check for the
+  configured workspace file (also allows the `preToolUse` deny hook to let
+  that write through).
+
+**Cadence** (unchanged)
 
 - `CONTINUAL_LEARNING_MIN_TURNS` (or legacy `CONTINUOUS_LEARNING_MIN_TURNS`)
 - `CONTINUAL_LEARNING_MIN_MINUTES` (or legacy `CONTINUOUS_LEARNING_MIN_MINUTES`)
@@ -57,7 +113,7 @@ Trial mode defaults (enabled in this plugin hook config):
 - `CONTINUAL_LEARNING_TRIAL_MIN_MINUTES` (or legacy `CONTINUOUS_LEARNING_TRIAL_MIN_MINUTES`)
 - `CONTINUAL_LEARNING_TRIAL_DURATION_MINUTES` (or legacy `CONTINUOUS_LEARNING_TRIAL_DURATION_MINUTES`)
 
-## Output format in AGENTS.md
+## Output format
 
 The memory updater writes only:
 
@@ -66,6 +122,18 @@ The memory updater writes only:
 
 Each item is a plain bullet point.
 
+## Migration
+
+Older versions stored cadence/index at
+`.cursor/hooks/state/continual-learning.json` and
+`continual-learning-index.json`. On first run the plugin copies them into the
+user-scoped state dir. Tracked originals are left in place; untracked
+originals are removed so they stop showing up in `git status`.
+
+To restore the old shared-file behavior, set
+`CONTINUAL_LEARNING_WORKSPACE_FILE` to the repo `AGENTS.md` and
+`CONTINUAL_LEARNING_ALLOW_SHARED=1`.
+
 ## License
 
 MIT

continual-learning/agents/agents-memory-updater.md

@@ -1,47 +1,75 @@
 ---
 name: agents-memory-updater
-description: Mine high-signal transcript deltas, update `AGENTS.md`, and keep the incremental transcript index in sync.
+description: Mine high-signal transcript deltas, update the user-scoped AGENTS.local.md memory file, and keep the incremental transcript index in sync.
 model: inherit
 ---
 
-# AGENTS.md memory updater
+# Memory updater
 
 Own the full memory update flow for continual learning.
 
 ## Trigger
 
 Use from `continual-learning` when transcript deltas may produce durable memory updates.
 
+## Target files
+
+**User memory file** (always written, personal, never committed):
+
+- Path from the followup message when present; otherwise
+  `$CONTINUAL_LEARNING_USER_FILE` or
+  `~/.cursor/projects/<slug>/AGENTS.local.md`.
+- `<slug>` is the absolute workspace path with the leading `/` stripped and
+  remaining `/` replaced by `-`.
+- Owns `## Learned User Preferences` and `## Learned Workspace Facts`.
+- Do **not** rename the workspace-facts heading to `(local)`.
+
+**Workspace memory file** (opt-in only):
+
+- `$CONTINUAL_LEARNING_WORKSPACE_FILE`. Unset by default — skip it.
+- Refuse the write when `git check-ignore -q` exits `1` unless
+  `CONTINUAL_LEARNING_ALLOW_SHARED=1`.
+
+**Never** write `## Learned User Preferences` or `## Learned Workspace Facts`
+into a repo-tracked `AGENTS.md` / `CLAUDE.md` / `GEMINI.md`.
+
+**Index**: `$CONTINUAL_LEARNING_STATE_DIR/index.json` if set; otherwise
+`~/.cursor/projects/<slug>/continual-learning/index.json`. Prefer the absolute
+index path embedded in the followup message.
+
 ## Workflow
 
-1. Read existing `AGENTS.md` first. If it does not exist, create it with only:
+1. Resolve target paths from the followup message when present.
+2. Read the existing user memory file. If it does not exist, create it with only:
    - `## Learned User Preferences`
    - `## Learned Workspace Facts`
-2. Load the incremental index if present.
-3. Inspect only transcript files under `~/.cursor/projects/<workspace-slug>/agent-transcripts/` that are new or have newer mtimes than the index.
-4. Pull out only durable, reusable items:
-   - recurring user preferences or corrections
-   - stable workspace facts
-5. Update `AGENTS.md` carefully:
+3. Load the incremental index.
+4. Inspect only transcript files under
+   `~/.cursor/projects/<workspace-slug>/agent-transcripts/` that are new or
+   have newer mtimes than the index.
+5. Extract only durable, reusable items.
+6. Update the user memory file:
    - update matching bullets in place
    - add only net-new bullets
    - deduplicate semantically similar bullets
-   - keep each learned section to at most 12 bullets
-6. Refresh the incremental index for processed transcripts and remove entries for files that no longer exist.
-7. If the merge produces no `AGENTS.md` changes, leave `AGENTS.md` unchanged but still refresh the index.
-8. If no meaningful updates exist, respond exactly: `No high-signal memory updates.`
+   - keep each section to at most 12 bullets
+7. Refresh the index for processed transcripts and remove entries for files
+   that no longer exist.
+8. If the merge produces no file changes, leave the memory file unchanged
+   but still refresh the index.
+9. If no meaningful updates exist, respond exactly:
+   `No high-signal memory updates.`
 
 ## Guardrails
 
 - Use plain bullet points only.
 - Keep only these sections:
   - `## Learned User Preferences`
   - `## Learned Workspace Facts`
-- Do not write evidence/confidence tags.
-- Do not write process instructions, rationale, or metadata blocks.
+- Do not write evidence/confidence tags, process instructions, rationale, or metadata.
 - Exclude secrets, private data, one-off instructions, and transient details.
 
 ## Output
 
-- Updated `AGENTS.md` and `.cursor/hooks/state/continual-learning-index.json` when needed
+- Updated the user memory file and the index when needed.
 - Otherwise exactly `No high-signal memory updates.`