@agenr/openclaw-plugin 0.13.0 → 0.13.2

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@agenr/openclaw-plugin",
-  "version": "0.13.0",
+  "version": "0.13.2",
   "description": "agenr memory plugin for OpenClaw — lightweight plugin-only package without the CLI",
   "license": "AGPL-3.0",
   "author": "agenr-ai",
@@ -25,9 +25,8 @@
   ],
   "dependencies": {
     "@libsql/client": "^0.17.2",
-    "@mariozechner/pi-ai": "^0.61.1",
-    "@sinclair/typebox": "^0.34.48",
-    "agenr": "0.13.0"
+    "@mariozechner/pi-ai": "^0.62.0",
+    "@sinclair/typebox": "^0.34.48"
   },
   "openclaw": {
     "extensions": [
@@ -40,4 +39,4 @@
     "memory",
     "plugin"
   ]
-}
+}

package/skills/skills/SKILL.md

@@ -1,77 +0,0 @@
----
-name: agenr
-description: Use when storing new knowledge (decisions, preferences, lessons, todos) or recalling context mid-session. The agenr plugin auto-injects memory at session start - this skill covers proactive store and on-demand recall.
----
-
-## agenr_store
-
-Use proactively for durable knowledge - decisions, preferences, lessons, and facts that future sessions need to make better decisions.
-
-**The future-session test:** Before storing, ask: "If I woke up in a fresh session and recalled this, would it change how I act?" If the answer is just "it tells me something that already happened and is done" - do not store it.
-
-**Store:**
-- Architecture decisions and design rationale
-- Workflow constraints and operational rules
-- Lessons learned from problems (recurring bugs, what worked, what did not)
-- User preferences and working-style observations
-- Durable facts about systems, environments, people, projects
-- Important open risks or unresolved hazards
-
-**Do not store:**
-- Version or release shipping events ("v1.2.3 shipped") - changelogs are the system of record
-- Issue or PR filing records ("filed #123, #124") - the issue tracker is the system of record
-- Phase plans or release packaging decisions - transient coordination artifacts, stale within a session
-- Progress snapshots ("3 of 5 tasks done") - stale within minutes
-- Prompt file paths or internal build logistics - look-up-able, not memory-worthy
-- Anything whose only value is duplicating files, the issue tracker, or git history
-
-Do not ask before storing - but do ask whether future-you needs it.
-
-Required: `type`, `content`, `importance`.
-Optional: `subject` (short label), `tags` (array), `scope` (`private`|`personal`|`public`), `project`, `platform`, `source`.
-
-Types: `fact | decision | preference | todo | lesson | event | relationship`
-Do not store secrets/credentials, temporary state, or verbatim conversation.
-
-### Importance calibration
-
-- **10**: Once-per-project permanent constraints. At most 1-2 per project lifetime.
-- **9**: Critical breaking changes or immediate cross-session decisions. At most 1 per significant session, often 0.
-- **8**: Things an active parallel session would act on right now. Fires a cross-session signal. Use conservatively.
-- **7**: Default. Project facts, decisions, preferences, milestones. No signal fired.
-- **6**: Routine dev observations (verified X, tests passing). Cap here unless the result is surprising.
-- **5**: Borderline. Only store if clearly durable beyond today.
-- **Importance is not recency.** Something that just happened is not automatically important. A shipping event is importance 5-6 at best unless it teaches a lesson. A recurring operational hazard is importance 7-8 because it affects future decisions.
-
-Entries at 7 are saved silently. Use 8+ only if other active sessions need to know NOW.
-
-### Confidence-aware extraction (OpenClaw transcripts)
-
-OpenClaw transcripts include `[user]` / `[assistant]` role labels. The extractor uses this signal:
-- Hedged or unverified assistant factual claims are tagged `unverified` and hard-capped at importance 5.
-- Tool-verified assistant claims follow normal importance rules.
-- User messages are never capped.
-
-This means: if you say something unverified, it will be stored at max importance 5. To store a fact at higher importance, verify it with a tool call first.
-
-## agenr_recall
-
-Use mid-session when you need context you don't already have. Session-start recall is handled automatically - do not call at turn 1 unless you need extra context beyond the injected summary.
-
-Parameters:
-- `query` (required): semantic search string
-- `limit`: max results (default 10)
-- `context`: `"default"` (semantic+vector) or `"session-start"` (fast bootstrap)
-- `since`: lower date bound - only entries newer than this (ISO or relative, e.g. `"7d"`, `"2026-01-01"`)
-- `until`: upper date bound - only entries older than this ceiling (e.g. `"7d"` = entries created before 7 days ago). Use with `since` for a date window.
-- `types`: comma-separated entry types (`fact,decision,preference,todo,lesson,event`)
-- `platform`: filter by platform (`openclaw`, `claude-code`, `codex`)
-- `project`: filter by project scope (pass `*` for all projects)
-
-## agenr_retire
-
-Soft-deletes an entry. Use when something is outdated, wrong, or superseded. Pass `entry_id` (from recall results) and optionally `reason` and `persist: true` to write to the retirements ledger.
-
-## agenr_extract
-
-Extracts structured knowledge entries from raw text without storing them. Useful for previewing what would be stored from a block of text before committing.