dsh-dejavu-memory 0.5.4 → 0.5.5

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/README.md CHANGED
@@ -4,21 +4,18 @@
4
4
 
5
5
  # dsh-dejavu-memory
6
6
 
7
- > **Renamed:** former npm/GitHub package `dsh-noc-memory` → **`dsh-dejavu-memory`**. Prefer this package; deprecate the old name when publishing.
8
-
9
-
10
7
  Connects DeepSeek Harness to **DejaVu**: session-start boot + daily briefing, plus memory read / search / create / update / delete, backed by your own DejaVu MCP server on Cloudflare.
11
8
 
12
9
  > Port of [pi-dejavu-memory](https://github.com/RetiredPhysicist/pi-plugins/tree/main/packages/pi-dejavu-memory) — same protocol, same tool names.
13
10
 
14
- There is also a **ChatGPT and Codex plugin**, [dejavu-memory](https://github.com/RetiredPhysicist/openai-plugins/tree/main/plugins/dejavu-memory) — the same memory server, but with an interactive knowledge graph of the tree inside the chat client. This plugin stays tool-only.
11
+ The [OpenAI/Codex plugin](https://github.com/RetiredPhysicist/openai-plugins/tree/main/plugins/dejavu-memory) is the only client that opens the interactive memory graph. dsh uses the same server through tools only, with no graph UI and no user-facing MCP setup.
15
12
 
16
13
  <p align="center">
17
- <img src="assets/readme/dejavu-memory-graph.webp" width="100%" alt="The DejaVu Memory plugin in Codex: the whole memory tree drawn as an interactive graph, with type colours and priority-sized nodes">
14
+ <img src="assets/readme/dejavu-memory-graph.webp" width="100%" alt="The OpenAI/Codex DejaVu Memory plugin showing the memory tree as an interactive graph">
18
15
  </p>
19
16
 
20
17
  <p align="center">
21
- <img src="assets/readme/dejavu-memory-inspector.webp" width="100%" alt="Selecting a node opens the inspector: the memory URI, its type and priority, and the body text, with its subtree highlighted">
18
+ <img src="assets/readme/dejavu-memory-inspector.webp" width="100%" alt="The OpenAI/Codex DejaVu Memory plugin inspector showing a node, its metadata, body text, and subtree">
22
19
  </p>
23
20
 
24
21
  [English](README.md) · [中文](README.zh.md)
@@ -40,7 +37,7 @@ There is also a **ChatGPT and Codex plugin**, [dejavu-memory](https://github.com
40
37
  dsh plugin --profile web add dsh-dejavu-memory
41
38
  ```
42
39
 
43
- Requires your own DejaVu server — deploy it to Cloudflare in minutes: [DejaVu](https://github.com/RealAlexandreAI/DejaVu).
40
+ Requires your own DejaVu MCP server on Cloudflare.
44
41
 
45
42
  ```yaml
46
43
  - id: dejavu
@@ -68,11 +65,9 @@ For a server behind Cloudflare Access (e.g. dejavu.example.com), use the **servi
68
65
  | `mcp_auth` | no | legacy; prefer `mcp_headers` for Cloudflare Access service token |
69
66
  | `mcp_headers` | no | extra headers merged into every MCP request (e.g. Cloudflare Access service token) |
70
67
 
71
- > **Upgrading from dsh-noc-memory:** package renamed to `dsh-dejavu-memory`. Remove the old plugin and re-add; point config `id` at `dejavu`.
72
-
73
- ## Why dejavu_* (not noc_*)?
68
+ ## Tool naming
74
69
 
75
- Some agents probe `read_mcp_resource` before reaching for a memory tool, wasting a round trip ([upstream issue #32](https://github.com/Dataojitori/nocturne_memory/issues/32)). Explicit `dejavu_boot` / `dejavu_read` naming in the tool list and boot-protocol prompt steers models straight to the right tool — no resource shim required.
70
+ The `dejavu_boot` / `dejavu_read` prefixes make the memory tools explicit in the tool list and in the boot protocol, so agents call them directly instead of probing for a generic resource first.
76
71
 
77
72
  ## License
78
73
 
@@ -80,7 +75,5 @@ MIT
80
75
 
81
76
  ## Related
82
77
 
83
- - [DejaVu](https://github.com/RealAlexandreAI/DejaVu) — the Cloudflare MCP memory server this plugin talks to
84
- - [dejavu-memory for ChatGPT and Codex](https://github.com/RetiredPhysicist/openai-plugins/tree/main/plugins/dejavu-memory) — same memory server, with an interactive knowledge graph
78
+ - [dejavu-memory for ChatGPT and Codex](https://github.com/RetiredPhysicist/openai-plugins/tree/main/plugins/dejavu-memory) — same memory server, with the interactive graph UI
85
79
  - [pi-dejavu-memory](https://github.com/RetiredPhysicist/pi-plugins/tree/main/packages/pi-dejavu-memory) — same memory tools for Pi
86
- - [DejaVu_memory](https://github.com/Dataojitori/DejaVu_memory) — upstream project
package/README.zh.md CHANGED
@@ -8,14 +8,14 @@
8
8
 
9
9
  > 由 [pi-dejavu-memory](https://github.com/RetiredPhysicist/pi-plugins/tree/main/packages/pi-dejavu-memory) 移植,协议与工具名完全一致。
10
10
 
11
- 另有 **ChatGPT / Codex 插件** [dejavu-memory](https://github.com/RetiredPhysicist/openai-plugins/tree/main/plugins/dejavu-memory),接同一个记忆服务器,但在聊天客户端里把记忆树画成可交互的知识图谱;本插件只提供工具。
11
+ 可交互的记忆图只由 [OpenAI/Codex 插件](https://github.com/RetiredPhysicist/openai-plugins/tree/main/plugins/dejavu-memory) 打开。dsh 通过同一服务的工具接入,没有图界面,也不需要让最终用户手工配 MCP。
12
12
 
13
13
  <p align="center">
14
- <img src="assets/readme/dejavu-memory-graph.webp" width="100%" alt="Codex 中的 DejaVu Memory 插件:整棵记忆树的可交互知识图谱,节点按类型着色、按优先级决定大小">
14
+ <img src="assets/readme/dejavu-memory-graph.webp" width="100%" alt="OpenAI/Codex 的 DejaVu Memory 插件:整棵记忆树的可交互图,节点按类型着色、按优先级决定大小">
15
15
  </p>
16
16
 
17
17
  <p align="center">
18
- <img src="assets/readme/dejavu-memory-inspector.webp" width="100%" alt="选中节点后展开详情面板:记忆 URI、类型与优先级、正文内容,并高亮其子树">
18
+ <img src="assets/readme/dejavu-memory-inspector.webp" width="100%" alt="OpenAI/Codex 的 DejaVu Memory 插件选中节点后的详情面板:记忆 URI、类型、优先级、正文和子树">
19
19
  </p>
20
20
 
21
21
  [English](README.md) · [中文](README.zh.md)
@@ -37,7 +37,7 @@
37
37
  dsh plugin --profile web add dsh-dejavu-memory
38
38
  ```
39
39
 
40
- 需要你能连到自己的 DejaVu(DejaVu)服务器——几分钟部署到 Cloudflare:[DejaVu](https://github.com/RealAlexandreAI/DejaVu)。
40
+ 需要你能连到自己的 DejaVu MCP 服务器。
41
41
 
42
42
  ```yaml
43
43
  - id: dejavu
@@ -64,11 +64,9 @@ dsh plugin --profile web add dsh-dejavu-memory
64
64
  | `mcp_url` | 是 | 你的 DejaVu MCP 端点(Streamable HTTP) |
65
65
  | `mcp_auth` | 否 | 遗留字段;优先用 `mcp_headers` 传 Cloudflare Access service token |
66
66
 
67
- > **从 dsh-noc-memory(≤0.1.x)升级:** 已改名为 `dsh-dejavu-memory`,工具名 `noc_*` → `dejavu_*`。删除旧插件后重新添加新包;更新提示词里所有旧工具引用。
67
+ ## 工具命名
68
68
 
69
- ## 为什么叫 dejavu_*(不叫 noc_*)?
70
-
71
- 部分模型会先探测 `read_mcp_resource` 再调用记忆工具,浪费一次往返([上游 issue #32](https://github.com/Dataojitori/nocturne_memory/issues/32))。工具列表与 boot 协议提示词里显式的 `dejavu_boot` / `dejavu_read` 命名,能把模型直接引导到正确工具——无需 resource 兼容层。
69
+ `dejavu_boot` / `dejavu_read` 前缀让记忆工具在工具列表和 boot 协议里含义明确,模型会直接调用,无需先探测通用 resource。
72
70
 
73
71
  ## License
74
72
 
@@ -76,7 +74,5 @@ MIT
76
74
 
77
75
  ## 相关
78
76
 
79
- - [DejaVu](https://github.com/RealAlexandreAI/DejaVu) — 本插件连接的 Cloudflare MCP 记忆服务器
80
- - [dejavu-memory(ChatGPT / Codex 版)](https://github.com/RetiredPhysicist/openai-plugins/tree/main/plugins/dejavu-memory) — 同一个记忆服务器,另有可交互的知识图谱
77
+ - [dejavu-memory(ChatGPT / Codex 版)](https://github.com/RetiredPhysicist/openai-plugins/tree/main/plugins/dejavu-memory) — 同一个记忆服务器,另有可交互的图界面
81
78
  - [pi-dejavu-memory](https://github.com/RetiredPhysicist/pi-plugins/tree/main/packages/pi-dejavu-memory) — 面向 Pi 的同一套记忆工具
82
- - [DejaVu_memory](https://github.com/Dataojitori/DejaVu_memory) — 上游项目
package/cordis.patch.yml CHANGED
@@ -6,7 +6,9 @@
6
6
  # name: 'dsh-dejavu-memory'
7
7
  # config:
8
8
  # mcp_url: http://localhost:PORT/mcp
9
- # mcp_auth: Bearer <your token>
9
+ # mcp_headers:
10
+ # CF-Access-Client-Id: <client id>
11
+ # CF-Access-Client-Secret: <client secret>
10
12
  - insert:
11
13
  - id: dejavu
12
14
  name: 'dsh-dejavu-memory'
package/dist/index.d.ts CHANGED
@@ -3,7 +3,7 @@ import z from '@deepseek-ai/schemastery';
3
3
  export declare const name = "dejavu-memory";
4
4
  export declare const inject: string[];
5
5
  export interface Config {
6
- /** Noc MCP server URL, e.g. http://localhost:PORT/mcp. Optional at load
6
+ /** DejaVu MCP server URL, e.g. http://localhost:PORT/mcp. Optional at load
7
7
  * time — a missing server surfaces as a setup hint on tool calls so the
8
8
  * plugin still loads in a bare profile. */
9
9
  mcp_url?: string;
@@ -20,13 +20,13 @@ export declare const Config: z<Config>;
20
20
  export declare function parseStreamResponse(text: string): any;
21
21
  export declare function extractText(data: any): string;
22
22
  /** Boot resources read via read_memory (boot + recent + triggers). */
23
- export declare const BOOT_URIS: readonly ["system://boot", "system://recent/5", "system://triggers"];
23
+ export declare const BOOT_URIS: readonly ['system://boot', 'system://recent/5', 'system://triggers'];
24
24
  /** MCP tool names as exposed by the DejaVu MCP server (must stay in sync with the server). */
25
25
  export declare const MCP_TOOLS: {
26
- readonly read: "read_memory";
27
- readonly search: "search_memory";
28
- readonly create: "create_memory";
29
- readonly update: "update_memory";
30
- readonly delete: "delete_memory";
26
+ readonly read: 'read_memory';
27
+ readonly search: 'search_memory';
28
+ readonly create: 'create_memory';
29
+ readonly update: 'update_memory';
30
+ readonly delete: 'delete_memory';
31
31
  };
32
32
  export declare function apply(ctx: Context, config: Config): void;
package/dist/index.js CHANGED
@@ -1,7 +1,7 @@
1
1
  // dsh-memory — DeepSeek Harness (Cordis) plugin.
2
2
  //
3
- // Noc Memory client: automated long-term memory for the agent,
4
- // backed by YOUR OWN Noc MCP server (mcp_url). Ported from
3
+ // DejaVu Memory client: automated long-term memory for the agent,
4
+ // backed by YOUR OWN DejaVu MCP server (mcp_url). Ported from
5
5
  // pi-dejavu-memory — same MCP protocol, same boot protocol, same tools.
6
6
  //
7
7
  // Tools: dejavu_boot (session-start memory load), dejavu_read,
@@ -13,8 +13,9 @@ import z from '@deepseek-ai/schemastery';
13
13
  import { defineTool } from '@deepseek-ai/dsh-tools';
14
14
  export const name = 'dejavu-memory';
15
15
  export const inject = ['tools', 'systemPrompt'];
16
+ const PLUGIN_VERSION = '0.5.5';
16
17
  export const Config = z.object({
17
- mcp_url: z.string().description('Noc MCP server URL, e.g. http://localhost:PORT/mcp'),
18
+ mcp_url: z.string().description('DejaVu MCP server URL, e.g. http://localhost:PORT/mcp'),
18
19
  mcp_auth: z.string().description('MCP auth token (Authorization header value, e.g. "Bearer xxx")'),
19
20
  mcp_headers: z.dict(z.string()).description('Extra headers merged into every MCP request (e.g. Cloudflare Access service token)'),
20
21
  protocol_version: z.string().description('MCP protocol version (default 2024-11-05)'),
@@ -89,7 +90,7 @@ function isMissingSession(parsed) {
89
90
  const message = String(parsed?.error?.message ?? '').toLowerCase();
90
91
  return code === -32600 || message.includes('session');
91
92
  }
92
- class NocClient {
93
+ class DejaVuClient {
93
94
  url;
94
95
  auth;
95
96
  protocolVersion;
@@ -130,7 +131,7 @@ class NocClient {
130
131
  const resp = await this.fetchRaw('initialize', {
131
132
  protocolVersion: this.protocolVersion,
132
133
  capabilities: {},
133
- clientInfo: { name: 'dsh-memory', version: '0.1.0' },
134
+ clientInfo: { name: 'dsh-dejavu-memory', version: PLUGIN_VERSION },
134
135
  });
135
136
  if (!resp.ok)
136
137
  return null;
@@ -175,21 +176,23 @@ export function apply(ctx, config) {
175
176
  ctx.systemPrompt.section({
176
177
  name: 'tool:dejavu-memory',
177
178
  order: 2960,
178
- text: 'You have long-term memory via the Noc MCP server. At the start of ' +
179
+ text: 'You have long-term memory via the DejaVu MCP server. At the start of ' +
179
180
  'substantial work call dejavu_boot to load system://boot, system://recent/5, ' +
180
181
  'system://triggers, and best-effort system://briefing; then ' +
181
182
  'read system://focus to resume active working trees (recent is a briefing ' +
182
- 'subset — no need to re-read it after boot); use dejavu_search before ' +
183
- 'answering from memory — describe what you need in natural language, not ' +
183
+ 'subset — no need to re-read it after boot). Use dejavu_search before ' +
184
+ 'answering from memory or when the task topic changes; describe what you need in natural language, not ' +
184
185
  'just keywords (semantic recall finds memories with no shared words); ' +
185
- 'persist valuable outcomes with dejavu_create; revise with dejavu_update; remove ' +
186
- 'dead nodes with dejavu_delete. ' +
186
+ 'read before writing, persist durable outcomes with dejavu_create, ' +
187
+ 'revise with dejavu_update, and remove dead nodes with dejavu_delete. ' +
188
+ 'Prefer dejavu_update when the topic already exists because URI is identity. ' +
187
189
  'Periodically (after many new memories or when you repeat a mistake) ' +
188
190
  'run a memory audit: dejavu_read system://diagnostic/dejavu, then fix what it ' +
189
191
  'flags — never-reaccessed high-priority memories (disclosure/placement), ' +
190
192
  'stale or cold candidates (dejavu_delete if dead, demote if niche), crowded ' +
191
- 'parents (regroup), contradictions (merge via dejavu_update). Always ' +
192
- 'dejavu_read a node in full before changing it.',
193
+ 'parents (regroup), contradictions (merge via dejavu_update). Never keep two conflicting versions. ' +
194
+ 'Always dejavu_read a node in full before changing or deleting it. ' +
195
+ 'The interactive memory graph belongs to the OpenAI/Codex plugin, not dsh.',
193
196
  });
194
197
  const resolveAuth = async () => config.mcp_auth ?? '';
195
198
  // One client per plugin instance: the MCP session (initialize handshake +
@@ -200,9 +203,9 @@ export function apply(ctx, config) {
200
203
  if (!cachedClient) {
201
204
  cachedClient = (async () => {
202
205
  if (!config.mcp_url) {
203
- throw new Error('Noc MCP server not configured — set mcp_url in the plugin config (e.g. http://localhost:PORT/mcp)');
206
+ throw new Error('DejaVu MCP server not configured — set mcp_url in the plugin config (e.g. http://localhost:PORT/mcp)');
204
207
  }
205
- return new NocClient(config.mcp_url, await resolveAuth(), config.protocol_version ?? '2024-11-05', config.mcp_headers ?? {});
208
+ return new DejaVuClient(config.mcp_url, await resolveAuth(), config.protocol_version ?? '2024-11-05', config.mcp_headers ?? {});
206
209
  })();
207
210
  }
208
211
  return cachedClient;
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "dsh-dejavu-memory",
3
- "version": "0.5.4",
3
+ "version": "0.5.5",
4
4
  "description": "DeepSeek Harness plugin: DejaVu memory client — automated long-term memory (boot/read/search/create/update/delete) backed by your own DejaVu MCP server.",
5
5
  "type": "module",
6
6
  "main": "dist/index.js",
@@ -35,9 +35,9 @@
35
35
  "@deepseek-ai/schemastery": "~3.18.5"
36
36
  },
37
37
  "devDependencies": {
38
- "@types/node": "^22.0.0",
39
- "tsx": "^4.19.0",
40
- "typescript": "^5.5.0"
38
+ "@types/node": "^26.6.5",
39
+ "tsx": "^4.23.15",
40
+ "typescript": "^7.0.2"
41
41
  },
42
42
  "dsh": {
43
43
  "manifestVersion": 1,
@@ -51,6 +51,6 @@
51
51
  "directory": "packages/dsh-dejavu-memory"
52
52
  },
53
53
  "engines": {
54
- "node": ">=20.0.0"
54
+ "node": ">=26"
55
55
  }
56
56
  }