transcendence-memory

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

What This Skill Does

该skill的功能

Provides self-hosted long-term memory for AI agents by connecting to the transcendence-memory-server backend.

Core capabilities:

Connect: complete authentication in one step with a connection token or manual configuration
Text memory: manage structured memories through lightweight CRUD endpoints
Multimodal RAG: upload documents (PDF, image, or Markdown) or raw text into the RAG-Anything pipeline, then ask natural-language questions and get LLM-generated answers
Container management: list and delete containers
Troubleshooting: diagnose connection and retrieval issues

通过连接transcendence-memory-server后端，为AI Agent提供自托管长期记忆能力。

核心能力：

连接：通过连接令牌一步完成身份验证，或手动配置连接
文本记忆：通过轻量级CRUD接口管理结构化记忆
多模态RAG：将文档（PDF、图片或Markdown）或原始文本上传到RAG-Anything流水线，之后即可用自然语言提问并获取LLM生成的答案
容器管理：列出和删除容器
故障排查：诊断连接和检索相关问题

Install

安装

bash

npx skills add https://github.com/leekkk2/transcendence-memory --skill transcendence-memory

Or inside a Claude Code session:

text

/plugin marketplace add leekkk2/transcendence-memory
/plugin install transcendence-memory

bash

npx skills add https://github.com/leekkk2/transcendence-memory --skill transcendence-memory

或在Claude Code会话中执行：

text

/plugin marketplace add leekkk2/transcendence-memory
/plugin install transcendence-memory

Principles

设计原则

Keep builtin memory: server-side memory augments the agent's builtin memory instead of replacing it
Zero dependency: no extra package installation is required; the agent can do everything with native tools such as curl, file I/O, and the Python standard library
Progressive loading: read
```
references/setup.md
```
during first-time setup, then this file is enough for day-to-day use

保留内置内存：服务端内存是Agent内置内存的补充，而非替代
零依赖：无需安装额外依赖包，Agent可通过curl、文件I/O、Python标准库等原生工具完成所有操作
渐进式加载：首次设置时读取
```
references/setup.md
```
，日常使用仅需参考本文件即可

Built-in Commands

内置命令

These commands can be invoked through

/transcendence-memory <command>

or the short form

/tm <command>

Command	Purpose	Example
`connect <token>`	Import a connection token and write local config	`/tm connect eyJlbmRw...`
`connect --manual`	Enter endpoint, api_key, and container manually	`/tm connect --manual`
`status`	Check connection status and server health	`/tm status`
`search <query>`	Run semantic search over memories	`/tm search architecture decision from the last deployment`
`remember <text>`	Store one memory quickly	`/tm remember Port conflicts caused the deployment failure`
`embed`	Rebuild the index for the current container	`/tm embed`
`query <question>`	Run a multimodal RAG query and get an LLM-generated answer	`/tm query What is the overall project architecture?`
`upload <file>`	Upload a file into the knowledge graph	`/tm upload ./design.pdf`
`containers`	List all containers	`/tm containers`
`batch <file.jsonl>`	Bulk import memories	`/tm batch memories.jsonl`
`auto on`	Enable automatic memory on git commits	`/tm auto on`
`auto off`	Disable automatic memory	`/tm auto off`
`auto status`	Show auto-memory configuration	`/tm auto status`

这些命令可通过

/transcendence-memory <命令>

调用，也可使用简写

/tm <命令>

：

命令	用途	示例
`connect <token>`	导入连接令牌并写入本地配置	`/tm connect eyJlbmRw...`
`connect --manual`	手动输入endpoint、api_key和container	`/tm connect --manual`
`status`	检查连接状态和服务端健康状态	`/tm status`
`search <query>`	对记忆执行语义搜索	`/tm search architecture decision from the last deployment`
`remember <text>`	快速存储单条记忆	`/tm remember Port conflicts caused the deployment failure`
`embed`	为当前容器重建索引	`/tm embed`
`query <question>`	执行多模态RAG查询并获取LLM生成的答案	`/tm query What is the overall project architecture?`
`upload <file>`	上传文件到知识图谱	`/tm upload ./design.pdf`
`containers`	列出所有容器	`/tm containers`
`batch <file.jsonl>`	批量导入记忆	`/tm batch memories.jsonl`
`auto on`	开启git提交自动记忆功能	`/tm auto on`
`auto off`	关闭自动记忆功能	`/tm auto off`
`auto status`	查看自动记忆配置	`/tm auto status`

Command:

connect

命令：

connect

Import a connection token or configure the connection manually.

Token mode (recommended):

bash

undefined

导入连接令牌或手动配置连接。

令牌模式（推荐）：

bash

undefined

Automatically run by the agent after it receives a token:

Agent收到令牌后自动执行：

TOKEN="$1" # base64 token provided by the user DECODED=$(echo "$TOKEN" | base64 -d) ENDPOINT=$(echo "$DECODED" | python3 -c "import sys,json; print(json.load(sys.stdin)['endpoint'])") API_KEY=$(echo "$DECODED" | python3 -c "import sys,json; print(json.load(sys.stdin)['api_key'])") CONTAINER=$(echo "$DECODED" | python3 -c "import sys,json; print(json.load(sys.stdin)['container'])")

mkdir -p ~/.transcendence-memory && chmod 700 ~/.transcendence-memory cat > ~/.transcendence-memory/config.toml << EOF [connection] endpoint = "$ENDPOINT" container = "$CONTAINER"

[auth] mode = "api_key" api_key = "$API_KEY" EOF chmod 600 ~/.transcendence-memory/config.toml

TOKEN="$1" # 用户提供的base64令牌 DECODED=$(echo "$TOKEN" | base64 -d) ENDPOINT=$(echo "$DECODED" | python3 -c "import sys,json; print(json.load(sys.stdin)['endpoint'])") API_KEY=$(echo "$DECODED" | python3 -c "import sys,json; print(json.load(sys.stdin)['api_key'])") CONTAINER=$(echo "$DECODED" | python3 -c "import sys,json; print(json.load(sys.stdin)['container'])")

mkdir -p ~/.transcendence-memory && chmod 700 ~/.transcendence-memory cat > ~/.transcendence-memory/config.toml << EOF [connection] endpoint = "$ENDPOINT" container = "$CONTAINER"

[auth] mode = "api_key" api_key = "$API_KEY" EOF chmod 600 ~/.transcendence-memory/config.toml

Verify the connection

验证连接

curl -sS "$ENDPOINT/health"


**Manual mode**: ask the user for `endpoint`, `api_key`, and `container`, then write `config.toml`.

curl -sS "$ENDPOINT/health"


**手动模式**：向用户询问`endpoint`、`api_key`和`container`，然后写入`config.toml`。

Command:

status

命令：

status

Check connection and server status:

bash

undefined

检查连接和服务端状态：

bash

undefined

Read local config

读取本地配置

Health check

健康检查

curl -sS "$ENDPOINT/health" | python3 -m json.tool

Authentication test

身份验证测试

curl -sS -X POST "$ENDPOINT/search"
-H "X-API-KEY: $API_KEY" -H "Content-Type: application/json"
-d "{"container":"$CONTAINER","query":"test","topk":1}"

undefined

curl -sS -X POST "$ENDPOINT/search"
-H "X-API-KEY: $API_KEY" -H "Content-Type: application/json"
-d "{"container":"$CONTAINER","query":"test","topk":1}"

undefined

Command:

search

命令：

search

bash

curl -sS -X POST "${ENDPOINT}/search" \
  -H "X-API-KEY: ${API_KEY}" -H "Content-Type: application/json" \
  -d "{\"container\":\"${CONTAINER}\",\"query\":\"$ARGUMENTS\",\"topk\":5}"

bash

curl -sS -X POST "${ENDPOINT}/search" \
  -H "X-API-KEY: ${API_KEY}" -H "Content-Type: application/json" \
  -d "{\"container\":\"${CONTAINER}\",\"query\":\"$ARGUMENTS\",\"topk\":5}"

Command:

remember

命令：

remember

Quickly store one memory with an auto-generated ID and automatic embedding:

bash

MEM_ID="mem-$(date +%s)"
curl -sS -X POST "${ENDPOINT}/ingest-memory/objects" \
  -H "X-API-KEY: ${API_KEY}" -H "Content-Type: application/json" \
  -d "{\"container\":\"${CONTAINER}\",\"objects\":[{\"id\":\"${MEM_ID}\",\"text\":\"$ARGUMENTS\",\"tags\":[]}],\"auto_embed\":true}"

快速存储单条记忆，自动生成ID并自动执行embedding：

bash

MEM_ID="mem-$(date +%s)"
curl -sS -X POST "${ENDPOINT}/ingest-memory/objects" \
  -H "X-API-KEY: ${API_KEY}" -H "Content-Type: application/json" \
  -d "{\"container\":\"${CONTAINER}\",\"objects\":[{\"id\":\"${MEM_ID}\",\"text\":\"$ARGUMENTS\",\"tags\":[]}],\"auto_embed\":true}"

Command:

query

命令：

query

Run a multimodal RAG query with knowledge graph retrieval plus LLM answer generation:

bash

curl -sS -X POST "${ENDPOINT}/query" \
  -H "X-API-KEY: ${API_KEY}" -H "Content-Type: application/json" \
  -d "{\"query\":\"$ARGUMENTS\",\"container\":\"${CONTAINER}\",\"mode\":\"hybrid\",\"top_k\":60}"

执行多模态RAG查询，结合知识图谱检索和LLM答案生成：

bash

curl -sS -X POST "${ENDPOINT}/query" \
  -H "X-API-KEY: ${API_KEY}" -H "Content-Type: application/json" \
  -d "{\"query\":\"$ARGUMENTS\",\"container\":\"${CONTAINER}\",\"mode\":\"hybrid\",\"top_k\":60}"

Command:

upload

命令：

upload

Upload a file into the knowledge graph:

bash

curl -sS -X POST "${ENDPOINT}/documents/upload" \
  -H "X-API-KEY: ${API_KEY}" \
  -F "file=@$1" \
  -F "container=${CONTAINER}"

上传文件到知识图谱：

bash

curl -sS -X POST "${ENDPOINT}/documents/upload" \
  -H "X-API-KEY: ${API_KEY}" \
  -F "file=@$1" \
  -F "container=${CONTAINER}"

Command:

batch

命令：

batch

Bulk ingest memories with the bundled script:

bash

python3 <skill-path>/scripts/batch-ingest.py \
  "${ENDPOINT}" "${API_KEY}" "${CONTAINER}" "$1"

使用捆绑脚本批量导入记忆：

bash

python3 <skill-path>/scripts/batch-ingest.py \
  "${ENDPOINT}" "${API_KEY}" "${CONTAINER}" "$1"

Quick Reference (for configured users)

快速参考（已配置用户适用）

Text Memories (lightweight path)

文本记忆（轻量路径）

bash

undefined

bash

undefined

Search memories

搜索记忆

curl -sS -X POST "${ENDPOINT}/search"
-H "X-API-KEY: ${API_KEY}" -H "Content-Type: application/json"
-d '{"container":"${CONTAINER}","query":"what you want to search for","topk":5}'

Store a memory

存储记忆

curl -sS -X POST "${ENDPOINT}/ingest-memory/objects"
-H "X-API-KEY: ${API_KEY}" -H "Content-Type: application/json"
-d '{"container":"${CONTAINER}","objects":[{"id":"mem-001","text":"content to store","tags":["tag1"]}]}'

Rebuild the index after storing a new memory

存储新记忆后重建索引

curl -sS -X POST "${ENDPOINT}/embed"
-H "X-API-KEY: ${API_KEY}" -H "Content-Type: application/json"
-d '{"container":"${CONTAINER}","background":false,"wait":true}'

Update a memory

更新记忆

curl -sS -X PUT "${ENDPOINT}/containers/${CONTAINER}/memories/mem-001"
-H "X-API-KEY: ${API_KEY}" -H "Content-Type: application/json"
-d '{"text":"updated content","tags":["new-tag"]}'

Delete a memory

删除记忆

curl -sS -X DELETE "${ENDPOINT}/containers/${CONTAINER}/memories/mem-001"
-H "X-API-KEY: ${API_KEY}"


> After updating or deleting a memory, run `/embed` to refresh the index.

curl -sS -X DELETE "${ENDPOINT}/containers/${CONTAINER}/memories/mem-001"
-H "X-API-KEY: ${API_KEY}"


> 更新或删除记忆后，执行`/embed`刷新索引。

Multimodal RAG (RAG-Anything pipeline)

多模态RAG（RAG-Anything流水线）

bash

undefined

bash

undefined

Ingest raw text into the knowledge graph

将原始文本导入知识图谱

curl -sS -X POST "${ENDPOINT}/documents/text"
-H "X-API-KEY: ${API_KEY}" -H "Content-Type: application/json"
-d '{"container":"${CONTAINER}","text":"long text to ingest...","description":"optional description"}'

Upload a file (PDF, image, or Markdown)

上传文件（PDF、图片或Markdown）

curl -sS -X POST "${ENDPOINT}/documents/upload"
-H "X-API-KEY: ${API_KEY}"
-F "file=@/path/to/document.pdf"
-F "container=${CONTAINER}"

Multimodal RAG query that returns an LLM-generated answer

多模态RAG查询，返回LLM生成的答案

curl -sS -X POST "${ENDPOINT}/query"
-H "X-API-KEY: ${API_KEY}" -H "Content-Type: application/json"
-d '{"query":"your question","container":"${CONTAINER}","mode":"hybrid","top_k":60}'

undefined

curl -sS -X POST "${ENDPOINT}/query"
-H "X-API-KEY: ${API_KEY}" -H "Content-Type: application/json"
-d '{"query":"your question","container":"${CONTAINER}","mode":"hybrid","top_k":60}'

undefined

Container Management

容器管理

bash

undefined

bash

undefined

List all containers

列出所有容器

curl -sS "${ENDPOINT}/containers" -H "X-API-KEY: ${API_KEY}"

Delete a container

删除容器

curl -sS -X DELETE "${ENDPOINT}/containers/${CONTAINER}"
-H "X-API-KEY: ${API_KEY}"

Health check

健康检查

curl -sS "${ENDPOINT}/health"


Variables are read from the local config file `~/.transcendence-memory/config.toml`.

curl -sS "${ENDPOINT}/health"


变量从本地配置文件`~/.transcendence-memory/config.toml`中读取。

First-Time Setup

首次设置

On first use, read

references/setup.md

to complete configuration.

The core flow has only two steps:

Get a connection token from the server (through the
```
/export-connection-token
```
endpoint or from an administrator)
Run
```
/tm connect <token>
```
to finish setup automatically

Or run

/tm connect --manual

and enter the values step by step.

After configuration is complete,
references/setup.md
no longer needs to be loaded into context.

首次使用时，请阅读

references/setup.md

完成配置。

核心流程仅两步：

从服务端获取连接令牌（通过
```
/export-connection-token
```
接口或向管理员索要）
执行
```
/tm connect <token>
```
自动完成设置

或执行

/tm connect --manual

逐步输入配置值。

配置完成后，无需再将
references/setup.md
加载到上下文。

API Reference

API参考

See

references/api-reference.md

for full request and response formats.

完整的请求和响应格式请查看

references/api-reference.md

。

Lightweight Path (text memory CRUD)

轻量路径（文本记忆CRUD）

Endpoint	Method	Purpose	Auth
`/health`	GET	Health check	Not required
`/search`	POST	Search memories	Required
`/embed`	POST	Rebuild index	Required
`/ingest-memory/objects`	POST	Write typed objects	Required
`/ingest-memory/contract`	GET	Inspect ingest semantic boundaries	Not required
`/ingest-structured`	POST	Ingest structured JSON	Required
`/containers/{container}/memories/{id}`	PUT	Update a memory	Required
`/containers/{container}/memories/{id}`	DELETE	Delete a memory	Required

接口	请求方法	用途	身份验证
`/health`	GET	健康检查	无需
`/search`	POST	搜索记忆	需要
`/embed`	POST	重建索引	需要
`/ingest-memory/objects`	POST	写入类型化对象	需要
`/ingest-memory/contract`	GET	检查摄入语义边界	无需
`/ingest-structured`	POST	摄入结构化JSON	需要
`/containers/{container}/memories/{id}`	PUT	更新记忆	需要
`/containers/{container}/memories/{id}`	DELETE	删除记忆	需要

Multimodal Path (RAG-Anything pipeline)

多模态路径（RAG-Anything流水线）

Endpoint	Method	Purpose	Auth
`/documents/text`	POST	Ingest text into the knowledge graph	Required
`/documents/upload`	POST	Upload PDF, image, or Markdown documents	Required
`/query`	POST	Run a multimodal RAG query	Required

接口	请求方法	用途	身份验证
`/documents/text`	POST	将文本摄入知识图谱	需要
`/documents/upload`	POST	上传PDF、图片或Markdown文档	需要
`/query`	POST	执行多模态RAG查询	需要

Administrative Endpoints

管理接口

Endpoint	Method	Purpose	Auth
`/containers`	GET	List containers	Required
`/containers/{name}`	DELETE	Delete a container	Required
`/export-connection-token`	GET	Export a connection token	Required
`/jobs/{pid}`	GET	Async job status	Required

Authentication methods:

X-API-KEY: <api-key>

Authorization: Bearer <api-key>

接口	请求方法	用途	身份验证
`/containers`	GET	列出容器	需要
`/containers/{name}`	DELETE	删除容器	需要
`/export-connection-token`	GET	导出连接令牌	需要
`/jobs/{pid}`	GET	异步任务状态	需要

身份验证方式：

X-API-KEY: <api-key>

或

Authorization: Bearer <api-key>

Architecture Overview

架构概览

See

references/ARCHITECTURE.md

text

Agent --HTTPS + API Key--> transcendence-memory-server
                            |-- FastAPI HTTP layer
                            |-- Container isolation
                            |-- Lightweight path: /search + /ingest + /embed
                            |   `-- Embedding -> LanceDB vector store
                            `-- Multimodal path: /documents + /query
                                `-- RAG-Anything -> knowledge graph -> LLM answer

请查看

references/ARCHITECTURE.md

。

text

Agent --HTTPS + API Key--> transcendence-memory-server
                            |-- FastAPI HTTP layer
                            |-- Container isolation
                            |-- Lightweight path: /search + /ingest + /embed
                            |   `-- Embedding -> LanceDB vector store
                            `-- Multimodal path: /documents + /query
                                `-- RAG-Anything -> knowledge graph -> LLM answer

Troubleshooting

故障排查

See

references/troubleshooting.md

Common quick checks:

Cannot connect: run
```
/tm status
```
or
```
curl -sS "${ENDPOINT}/health"
```
401/403: verify that the API key is correct
Search returns empty: run
```
/tm embed
```
first to rebuild the index
Search returns 200 but the body contains an error: treat it as a failure and inspect the server logs
Document upload fails: verify file type and size (supported types include PDF, image, and Markdown)
Query returns empty: make sure content has been ingested through
```
/documents/text
```
or
```
/documents/upload
```
Updates or deletes do not appear in search: run
```
/tm embed
```
to refresh the index

请查看

references/troubleshooting.md

。

常见快速检查项：

无法连接：执行
```
/tm status
```
或
```
curl -sS "${ENDPOINT}/health"
```
401/403错误：验证API key是否正确
搜索返回空结果：先执行
```
/tm embed
```
重建索引
搜索返回200但响应体包含错误：视为请求失败，检查服务端日志
文档上传失败：验证文件类型和大小（支持的类型包括PDF、图片和Markdown）
查询返回空结果：确认已通过
```
/documents/text
```
或
```
/documents/upload
```
摄入了内容
更新或删除的内容未在搜索中体现：执行
```
/tm embed
```
刷新索引

Batch and Async Operations

批量和异步操作

Bulk Ingest (large memory sets)

批量摄入（大量记忆集）

When you need to ingest dozens to hundreds of memories:

bash

undefined

当需要摄入几十到上百条记忆时：

bash

undefined

Prepare a JSONL file, one JSON object per line

准备JSONL文件，每行一个JSON对象

{"id":"mem-001","text":"memory content","tags":["tag1"]}

{"id":"mem-002","text":"another memory","source":"telegram"}

/tm batch memories.jsonl

Or call the script directly:

或直接调用脚本：

python3 <skill-path>/scripts/batch-ingest.py
"${ENDPOINT}" "${API_KEY}" "${CONTAINER}" memories.jsonl


The script automatically batches input in groups of 50, retries failed requests, and prints progress output. It has zero external dependencies and uses only the Python standard library.

python3 <skill-path>/scripts/batch-ingest.py
"${ENDPOINT}" "${API_KEY}" "${CONTAINER}" memories.jsonl


该脚本自动将输入按50条分组，重试失败的请求，并打印进度输出。它没有外部依赖，仅使用Python标准库。

Async Tasks

异步任务

/embed

and

/documents/upload

support async mode:

bash

undefined

/embed

和

/documents/upload

支持异步模式：

bash

undefined

Submit an index rebuild asynchronously

异步提交索引重建任务

curl -sS -X POST "${ENDPOINT}/embed"
-H "X-API-KEY: ${API_KEY}" -H "Content-Type: application/json"
-d '{"container":"${CONTAINER}","background":true}'

Check async task status

查看异步任务状态

curl -sS "${ENDPOINT}/jobs/${PID}" -H "X-API-KEY: ${API_KEY}"

undefined

curl -sS "${ENDPOINT}/jobs/${PID}" -H "X-API-KEY: ${API_KEY}"

undefined

Choosing an Operation Mode

操作模式选择

Scenario	Recommended approach
Health checks, single searches, or a few memory writes	Built-in `/tm` commands
Bulk ingest of dozens to hundreds of memories	`/tm batch file.jsonl`
Rebuilding a large container index	`/tm embed` or async mode
Adding documents to the knowledge graph	`/tm upload file.pdf` or `/documents/text`
Asking for an LLM-synthesized answer	`/tm query your question`

场景	推荐方案
健康检查、单次搜索或少量记忆写入	内置 `/tm` 命令
批量摄入几十到上百条记忆	`/tm batch file.jsonl`
重建大型容器索引	`/tm embed` 或异步模式
向知识图谱添加文档	`/tm upload file.pdf` 或 `/documents/text`
请求LLM合成的答案	`/tm query your question`

Command:

auto

命令：

auto

Enable, disable, or check automatic memory management.

Enable — creates a marker file so hooks auto-store commit summaries:

bash

mkdir -p ~/.transcendence-memory
touch ~/.transcendence-memory/auto-memory.enabled
echo "Automatic memory enabled. Git commit summaries will be stored automatically."

Disable — removes the marker file:

bash

rm -f ~/.transcendence-memory/auto-memory.enabled
echo "Automatic memory disabled."

Status — check current state:

bash

if [ -f ~/.transcendence-memory/auto-memory.enabled ]; then
  echo "Automatic memory: ENABLED"
else
  echo "Automatic memory: DISABLED"
fi
if [ -f ~/.transcendence-memory/config.toml ]; then
  ENDPOINT=$(grep '^endpoint' ~/.transcendence-memory/config.toml | sed 's/.*= *"//' | sed 's/".*//')
  CONTAINER=$(grep '^container' ~/.transcendence-memory/config.toml | sed 's/.*= *"//' | sed 's/".*//')
  echo "Endpoint: ${ENDPOINT}"
  echo "Container: ${CONTAINER}"
else
  echo "Not connected. Run /tm connect first."
fi

启用、禁用或检查自动记忆管理功能。

启用 — 创建标记文件，让钩子自动存储提交摘要：

bash

mkdir -p ~/.transcendence-memory
touch ~/.transcendence-memory/auto-memory.enabled
echo "Automatic memory enabled. Git commit summaries will be stored automatically."

禁用 — 删除标记文件：

bash

rm -f ~/.transcendence-memory/auto-memory.enabled
echo "Automatic memory disabled."

状态 — 检查当前状态：

bash

if [ -f ~/.transcendence-memory/auto-memory.enabled ]; then
  echo "Automatic memory: ENABLED"
else
  echo "Automatic memory: DISABLED"
fi
if [ -f ~/.transcendence-memory/config.toml ]; then
  ENDPOINT=$(grep '^endpoint' ~/.transcendence-memory/config.toml | sed 's/.*= *"//' | sed 's/".*//')
  CONTAINER=$(grep '^container' ~/.transcendence-memory/config.toml | sed 's/.*= *"//' | sed 's/".*//')
  echo "Endpoint: ${ENDPOINT}"
  echo "Container: ${CONTAINER}"
else
  echo "Not connected. Run /tm connect first."
fi

Automatic Memory

自动记忆

When enabled, transcendence-memory automatically stores a memory after every git commit, merge, cherry-pick, or rebase. This is powered by lifecycle hooks that integrate with the host AI coding CLI.

启用后，transcendence-memory会在每次git commit、merge、cherry-pick或rebase后自动存储记忆。该功能由与宿主AI编码CLI集成的生命周期钩子提供支持。

How it works

工作原理

A SessionStart hook fires when a new session begins. It checks the connection status and tells the agent whether auto-memory is enabled.
A PostToolUse hook fires after every shell command. If the command was a git commit and auto-memory is enabled, the agent is instructed to store a one-line commit summary as a memory tagged
```
auto-commit
```
.

新会话开始时触发SessionStart钩子，检查连接状态并告知Agent自动记忆是否启用。
每次shell命令执行后触发PostToolUse钩子，如果命令是git commit且自动记忆已启用，将指示Agent存储一行提交摘要作为记忆，标记为
```
auto-commit
```
标签。

Enable / disable

启用/禁用

text

/tm auto on       # enable auto-memory
/tm auto off      # disable auto-memory
/tm auto status   # check current configuration

text

/tm auto on       # 启用自动记忆
/tm auto off      # 禁用自动记忆
/tm auto status   # 查看当前配置

What gets stored

存储内容

Each auto-commit memory follows this format:

[commit abc1234] fix: resolve port conflict in docker-compose | files: M docker-compose.yml, M .env.example

All auto-commit memories are tagged

auto-commit

for easy filtering:

text

/tm search auto-commit

每条自动提交记忆遵循以下格式：

[commit abc1234] fix: resolve port conflict in docker-compose | files: M docker-compose.yml, M .env.example

所有自动提交记忆都带有

auto-commit

标签，便于筛选：

text

/tm search auto-commit

Platform Support

平台支持

The hooks system is designed to work across multiple AI coding CLIs. The plugin ships pre-built hook configs for supported platforms.

钩子系统设计为适配多个AI编码CLI，插件附带了支持平台的预构建钩子配置。

Claude Code (primary)

Claude Code（主要支持）

Hooks are registered in

hooks/hooks.json

and activated automatically when the plugin is installed via

/plugin install

钩子在

hooks/hooks.json

中注册，通过

/plugin install

安装插件时自动激活。

Cursor

Uses

hooks/hooks-cursor.json

with camelCase event names (

sessionStart

postToolUse

使用

hooks/hooks-cursor.json

，采用驼峰命名的事件名（

sessionStart

、

postToolUse

）。

Other platforms

其他平台

The multi-platform adapter (

hooks/adapter.py

) normalizes hook input from:

Platform	Event format	Detection
Claude Code	`hook_event_name` + `tool_name`	`CLAUDE_PLUGIN_ROOT` env
Cursor	Same JSON schema	`CURSOR_PLUGIN_ROOT` env
Gemini CLI	`AfterTool` + `matcher`	`matcher` field in JSON
Windsurf	`post-tool-use` + `tool` + `arguments`	`arguments` field in JSON
Vibe CLI	`post-tool-call` + `tool` + `input`	`input` field in JSON
Cline / Roo Code	`tool_name` or `tool` + JSON stdin/stdout	JSON structure detection
Copilot CLI	Claude Code compatible	`COPILOT_CLI` env
Augment Code	Claude Code compatible	Fallback to Claude format

For platforms without native hook support, add transcendence-memory instructions to the platform's rules file (e.g.,

.cursorrules

AGENTS.md

.clinerules/

多平台适配器（

hooks/adapter.py

）会标准化来自以下平台的钩子输入：

平台	事件格式	检测方式
Claude Code	`hook_event_name` + `tool_name`	`CLAUDE_PLUGIN_ROOT` 环境变量
Cursor	相同JSON schema	`CURSOR_PLUGIN_ROOT` 环境变量
Gemini CLI	`AfterTool` + `matcher`	JSON中的 `matcher` 字段
Windsurf	`post-tool-use` + `tool` + `arguments`	JSON中的 `arguments` 字段
Vibe CLI	`post-tool-call` + `tool` + `input`	JSON中的 `input` 字段
Cline / Roo Code	`tool_name` 或 `tool` + JSON标准输入/输出	JSON结构检测
Copilot CLI	兼容Claude Code	`COPILOT_CLI` 环境变量
Augment Code	兼容Claude Code	fallback到Claude格式

对于没有原生钩子支持的平台，将transcendence-memory说明添加到平台的规则文件中即可（例如

.cursorrules

、

AGENTS.md

、

.clinerules/

）。

Files in This Skill

该skill包含的文件

File	Purpose	When to load
`references/setup.md`	First-time setup guide	First use only
`references/api-reference.md`	Complete API reference	When API details are needed
`references/ARCHITECTURE.md`	Architecture and data flow	When understanding the system
`references/OPERATIONS.md`	Operational verification and acceptance	During deployment verification
`references/troubleshooting.md`	Troubleshooting guide	When something goes wrong
`references/templates/config.toml.template`	Config file template	During first-time setup
`scripts/batch-ingest.py`	Bulk ingest script	For large memory imports

文件	用途	加载时机
`references/setup.md`	首次设置指南	仅首次使用时
`references/api-reference.md`	完整API参考	需要API细节时
`references/ARCHITECTURE.md`	架构和数据流	了解系统原理时
`references/OPERATIONS.md`	运行验证和验收	部署验证阶段
`references/troubleshooting.md`	故障排查指南	出现问题时
`references/templates/config.toml.template`	配置文件模板	首次设置期间
`scripts/batch-ingest.py`	批量摄入脚本	大量记忆导入时

When NOT to Use

不适用场景

Deploying the backend service -> use the
```
transcendence-memory-server
```
repository
Managing Docker, systemd, or Nginx -> use the
```
transcendence-memory-server
```
repository
Troubleshooting server-side problems such as 5xx errors, storage issues, or logs -> use the
```
transcendence-memory-server
```
repository
Configuring Embedding, LLM, or VLM models -> this is a server-side concern and does not need to be handled by the skill

部署后端服务 → 请使用
```
transcendence-memory-server
```
仓库
管理Docker、systemd或Nginx → 请使用
```
transcendence-memory-server
```
仓库
排查服务端问题，例如5xx错误、存储问题或日志 → 请使用
```
transcendence-memory-server
```
仓库
配置Embedding、LLM或VLM模型 → 这是服务端配置项，无需该skill处理

transcendence-memory

Original

Translation

What This Skill Does

该skill的功能

Install

安装

Principles

设计原则

Built-in Commands

内置命令

Command: connect

命令：connect

Automatically run by the agent after it receives a token:

Agent收到令牌后自动执行：

Verify the connection

验证连接

Command: status

命令：status

Read local config

读取本地配置

Health check

健康检查

Authentication test

身份验证测试

Command: search

命令：search

Command: remember

命令：remember

Command: query

命令：query

Command: upload

命令：upload

Command: batch

命令：batch

Quick Reference (for configured users)

快速参考（已配置用户适用）

Text Memories (lightweight path)

文本记忆（轻量路径）

Search memories

搜索记忆

Store a memory

存储记忆

Rebuild the index after storing a new memory

存储新记忆后重建索引

Update a memory

更新记忆

Delete a memory

删除记忆

Multimodal RAG (RAG-Anything pipeline)

多模态RAG（RAG-Anything流水线）

Ingest raw text into the knowledge graph

将原始文本导入知识图谱

Upload a file (PDF, image, or Markdown)

上传文件（PDF、图片或Markdown）

Multimodal RAG query that returns an LLM-generated answer

多模态RAG查询，返回LLM生成的答案

Container Management

容器管理

List all containers

列出所有容器

Delete a container

删除容器

Health check

健康检查

First-Time Setup

首次设置

API Reference

API参考

Lightweight Path (text memory CRUD)

轻量路径（文本记忆CRUD）

Multimodal Path (RAG-Anything pipeline)

多模态路径（RAG-Anything流水线）

Administrative Endpoints

管理接口

Architecture Overview

架构概览

Troubleshooting

故障排查

Batch and Async Operations

Command:
`connect`

命令：
`connect`

Command:
`status`

命令：
`status`

Command:
`search`

命令：
`search`

Command:
`remember`

命令：
`remember`

Command:
`query`

命令：
`query`

Command:
`upload`

命令：
`upload`

Command:
`batch`

命令：
`batch`

Command:
`auto`

命令：
`auto`