eklavya 1.24.2 → 1.25.1

package/dist/user-skill/eklavya/SKILL.md

@@ -13,7 +13,8 @@ and tracks mastery with spaced repetition.
 
 Both halves are local by default: the summariser and the search index run on
 this machine and nothing leaves it unless `providers.observer` has been
-configured, which is an explicit choice with its own key.
+configured — an explicit choice that sends session batches to a Claude model
+through `claude -p`, on the developer's own Claude subscription.
 
 The two halves are switched separately, and their names now say so.
 `quiz.enabled: false` stops the questions and leaves memory recording;
@@ -54,7 +55,7 @@ above resolved to.
 In a Claude Code session with the Eklavya plugin loaded, `get_config`,
 `set_config`, `get_learner_profile`, `get_concept_graph` and `get_gate_status`
 are available as tools. Use them: they validate the values, they report which
-settings a repo is overriding, and they need no subprocess.
+settings this project is overriding, and they need no subprocess.
 
 The CLI is the fallback for everywhere else — a plain terminal, Cursor, a
 session where the plugin is not enabled. The two write the same files, so it
@@ -130,9 +131,18 @@ they set it back.
 eklavya config set focus learn --topic "database indexing"
 ```
 
-Repo config wins over global. When someone's personal setting has stopped
-applying, that is why — `config get` prints both paths, and `eklavya doctor` or
-`get_config` names the keys the repo is overriding.
+Project settings win over global ones. When someone's global setting has
+stopped applying, that is why — `config get` prints both paths, and `eklavya
+doctor` or `get_config` names the keys this project is overriding. The project
+file lives under `~/.eklavya/projects/`, never in the repository.
+
+`providers` is the exception: global only. A project file that sets it is
+ignored (`config get` and `get_config`'s `ignored_in_project` say so), and
+`set_config` at project scope returns `global_only`. Set it globally or not at
+all. `set_config` can also return `unreadable_config` — the config file does not
+parse and was left untouched; name the file and let them fix it — and
+`project_collision`, when another checkout's settings already sit in the file
+this one would use.
 
 Other keys, same `config set` shape: `pass_threshold`,
 `max_questions_per_task`, `min_minutes_between_quizzes`,
@@ -193,9 +203,14 @@ questions, or seems to have switched itself off. It never fails loudly — every
 hook exits successfully by design, so it can't break a session — which means a
 broken install and a quiet one look the same from the outside.
 
-Run `eklavya doctor`. Its first four lines are the install itself: `runtime`,
-`driver`, `plugin`, `skill`. If any says `FAILED`, it exits non-zero and prints
-the fix.
+Run `eklavya doctor`. Its first lines are the install itself: `runtime`,
+`driver`, `plugin`, `versions`, `skill`. If any says `FAILED`, it exits non-zero
+and prints the fix. Later rows also flag a config file that does not parse and a
+terminal commit gate missing `jq` or `sqlite3`.
+
+If session start printed `Eklavya paused · can't open its database · run:
+eklavya doctor` (or the variant naming a SQLite module and Node version
+mismatch), the database exists but would not open — `doctor` names why.
 
 The fix is almost always `eklavya install`. It is idempotent — it reinstalls the
 runtime, re-copies the plugin and rewrites the registration, and never touches
@@ -227,20 +242,29 @@ It opens the browser itself, so do not tell them to click the URL. Use
 `--no-open` when they only asked *where* the dashboard is, or when the session
 is on a machine with no desktop.
 
-Six sections, each deep-linkable — hand back the one that answers what was
-actually asked rather than the bare root:
+Two workflows, Learning and Memory, each with its own Dashboard and every page
+deep-linkable — hand back the one that answers what was actually asked rather
+than the bare root. Add `?project=<absolute repo path>` to open it scoped to one
+project; the ids are the ones `/api/projects` lists.
 
 | Ask | Link |
 |---|---|
-| how am I doing, streaks, activity | `/#/overview` |
-| what am I learning, search a concept | `/#/concepts` (or `/#/concepts/due`, `/mastered`, `/unseen`), `/#/concept/<slug>` for one |
-| what is due, what is scheduled, what I skipped | `/#/review`, `/#/review/upcoming`, `/#/review/skipped` |
-| what did that session teach me | `/#/sessions` |
-| what does this project remember | `/#/memory`, `/#/entry/<id>` for one observation and its evidence |
-| what has recall actually saved | `/#/reuse` |
-| is capture healthy | `/#/health` |
-| how hard is this repo allowed to get | `/#/projects` |
-| where are the gaps | `/#/domains` |
+| how am I doing, streaks, activity | `/#/learning/dashboard` |
+| what am I learning, search a concept | `/#/learning/concepts` (or `/#/learning/concepts/due`, `/mastered`, `/unseen`), `/#/learning/concept/<slug>` for one |
+| what is due, what is scheduled, what I skipped | `/#/learning/review`, `/#/learning/review/upcoming`, `/#/learning/review/skipped` |
+| what did that session teach me | `/#/learning/sessions`, `/#/learning/session/<id>` |
+| how hard is this repo allowed to get, which projects have no answers yet | `/#/learning/projects` |
+| where are the gaps | `/#/learning/domains` |
+| what does memory hold, at a glance | `/#/memory/dashboard` |
+| what does this project remember | `/#/memory/timeline` (or `/#/memory/timeline/<type>`), `/#/memory/entry/<id>` for one observation and its evidence |
+| which sessions captured anything | `/#/memory/sessions` |
+| which projects has memory recorded | `/#/memory/projects` |
+| what has recall actually saved | `/#/memory/reuse` |
+| is capture healthy | `/#/memory/health` |
+
+The older single-workflow links (`/#/overview`, `/#/concepts`, `/#/memory`,
+`/#/entry/<id>` and the rest) still open the right page, but hand back the
+canonical form.
 
 ## When the plugin is loaded, point at the commands
 
@@ -278,5 +302,6 @@ skill, not this one.
   entry and its vectors are gone. Confirm before the hard one, and never offer
   it as a tidying-up suggestion.
 - Do not configure `providers.observer` on the user's behalf. It is the one
-  setting that sends this machine's work to an API, and it needs their explicit
-  yes and their own key in an environment variable.
+  setting that sends this machine's work off it — to a Claude model, on their
+  subscription — and it needs their explicit yes. It goes in the global config
+  only.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "eklavya",
-  "version": "1.24.2",
+  "version": "1.25.1",
   "description": "Learn while your agent works — local knowledge graph, spaced repetition, and quiz gating for agent-assisted development.",
   "license": "MIT",
   "repository": {
@@ -54,6 +54,7 @@
   "devDependencies": {
     "@types/better-sqlite3": "^7.6.11",
     "@types/node": "^22.0.0",
+    "playwright-core": "^1.63.0",
     "tsx": "^4.19.0",
     "typescript": "^5.6.0",
     "vitest": "^4.1.11"