@lmzhen/dsh-tool-memory 0.32.4 → 0.32.6

package/README.md

@@ -1,18 +1,19 @@
 # @lmzhen/dsh-tool-memory
 
-Model-facing memory tool and prompt context: it registers the `memory` tool, one fixed
-guidance section and the runtime snapshot of the current memory entries. It owns no store —
-the memory files and their provider do.
+Model-facing `memory` tool plus durable-memory guidance and snapshot sections for the system prompt.
 
-## Model surface
+## Tool contract
 
-- **Model-visible:** the `memory` tool schema, one fixed guidance section, and a runtime snapshot containing the current memory entries; the snapshot scales with the stored entries and is absent when memory is empty.
-- **Prompt prefix / KV cache:** guidance text is prefix-stable; the runtime snapshot is replaced after successful memory writes and is otherwise unchanged between requests; family rules: `packages/README.md` §"Model-visible prompt prefix and the KV cache".
-- **Mount it?** yes — the `tool-memory` row, carried by the `evolution-all` and one-click `evolution-preset` bundles and by the Evolution agent preset delta.
+`target` selects `memory` for agent notes or `user` for stable user-profile facts. Actions are `add`, `replace`, and `remove`. `facts` and `content` are aliases; supplying different non-blank values for both is rejected.
 
-## Known limitations
+Use one `operations` array for a batch. The batch is evaluated against the final result so removal/replacement can free budget before additions. Add requires content; replace requires matching old text and new content; remove requires matching old text. Empty batches and malformed operations are rejected before approval.
 
-- No known durable consumer gaps at this time. Runtime contracts are covered by package and boundary tests.
-- The pre-approval required-field pre-check and the store's own rejection text are hand-copied twins: the check runs on BOTH paths (every `operations[]` element and the normalized single operation) BEFORE the approval gate, so an approval-enabled deployment never stages a write the approved replay must then fail. Keeping the two texts in step is still manual; unifying them is deferred.
+## Defaults and boundaries
 
-**Runtime invariant:** No companion is published. The platform auto-assembles nothing and the family mounts no `<pkg>/invariant` cordis row, so a companion here would never execute (v37 S2.1 / I-3).
+- `memoryEnabled`: default `true`; disabling removes the model tool and prompt sections while retaining an explicit refusal runner for previously staged writes.
+- `entryPreviewChars`: default 200 characters per entry in the tool response; this does not truncate stored content.
+- `memoryChars` and `userChars`: enforced by the selected provider, not by this tool; see [memory-files](../memory-files/README.md).
+
+When ordinary approval is enabled, a write can return `pending_id` instead of executing immediately. An effective platform session policy of `never` bypasses staging as designed.
+
+Guidance and snapshot injection require the optional `systemPrompt` service. Without it, the write tool still works but memory context is not injected. Successful writes through the memory registry refresh the snapshot, including graph and background-review writes.

package/lib/index.js

@@ -1,7 +1,7 @@
 import { effectiveSessionPolicy } from "@lmzhen/dsh-evolution-approval";
 import z from "@deepseek-ai/schemastery";
 import { defineTool } from "@deepseek-ai/dsh-tools";
-import { MEMORY_GUIDANCE_SECTION_ORDER, clampedNumber, payloadText, resolveExecOrigins, unwrapLive } from "@lmzhen/dsh-evolution-core";
+import { MEMORY_GUIDANCE_SECTION_ORDER, clampedNumber, payloadText, registerOwnerSettings, resolveExecOrigins, unwrapLive } from "@lmzhen/dsh-evolution-core";
 //#region lib/types/index.js
 /**
 * Model-facing memory tool and runtime-context memory snapshot.
@@ -32,6 +32,7 @@ const Config = z.object({
 	entryPreviewChars: z.number().min(1).default(DEFAULT_ENTRY_PREVIEW_CHARS).volatile()
 });
 async function apply(ctx, rawConfig = {}) {
+	registerOwnerSettings(ctx, "tool-memory", Config);
 	ctx.inject(["settings"], (injected) => {
 		const settings = injected.settings;
 		if (typeof settings?.configure !== "function") return;

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@lmzhen/dsh-tool-memory",
   "description": "Model-facing memory tool and prompt context (community build)",
-  "version": "0.32.4",
+  "version": "0.32.6",
   "publishConfig": {
     "access": "public"
   },
@@ -27,22 +27,22 @@
   "license": "MIT",
   "dependencies": {
     "@deepseek-ai/schemastery": "^3.18.5",
-    "@lmzhen/dsh-evolution-approval": "^0.32.4",
-    "@lmzhen/dsh-evolution-core": "^0.32.4"
+    "@lmzhen/dsh-evolution-approval": "^0.32.6",
+    "@lmzhen/dsh-evolution-core": "^0.32.6"
   },
   "peerDependencies": {
     "@deepseek-ai/cordis": "^4.0.5",
     "@deepseek-ai/dsh-tools": "^0.2.1-rc.2",
     "@deepseek-ai/dsh-system-prompt": "^0.2.1-rc.2",
-    "@lmzhen/dsh-memory": "^0.32.4"
+    "@lmzhen/dsh-memory": "^0.32.6"
   },
   "devDependencies": {
     "@deepseek-ai/dsh-tools": "^0.2.1-rc.2",
     "@deepseek-ai/dsh-system-prompt": "^0.2.1-rc.2",
     "@deepseek-ai/dsh-agent-loop-testkit": "^0.2.1-rc.2",
-    "@lmzhen/dsh-evolution-approval": "^0.32.4",
-    "@lmzhen/dsh-memory": "^0.32.4",
-    "@lmzhen/dsh-memory-files": "^0.32.4"
+    "@lmzhen/dsh-evolution-approval": "^0.32.6",
+    "@lmzhen/dsh-memory": "^0.32.6",
+    "@lmzhen/dsh-memory-files": "^0.32.6"
   },
   "engines": {
     "dsh": ">=0.2.1-rc.2 <0.2.2-0"