Memgentic 会话排查

其他社区

作者仓库星标 0
作者仓库 skills-registry

解读按原文结构重写，命令、链接、术语均保留；右侧可核对作者原始 SKILL.md

方法与流程

Capture Methods：Memories enter the system through: autodaemon — File watcher captures conversations automatically in the background. mcptool — Stored via MCP tool calls during AI sessions.

适用与边界

When to Use：Search memory before doing work, not after: Starting a new task — Check for prior decisions, context, or related work. Making architectural decisions — Check if this was discussed before.
当前原文没有单列不适用场景或限制。

原文中的明确线索

要点：「before」、「Starting a new task」、「Making architectural decisions」、「Debugging」、「User asks about history」、「Encountering unfamiliar code」、「Setting up or configuring something」、「Start compact」
文件与命令：memgenticrecall、memgenticremember、memgenticsearch、memgenticrecent、memgenticbriefing、memgenticsources、memgenticconfiguresession、memgenticstats

流狐整理：以上内容来自当前 SKILL.md 的章节与原词；未补写作者没有声明的工具、兼容性或能力。

Memgentic — Universal AI Memory Layer

Memgentic captures knowledge from all your AI tools and makes it searchable. Every memory carries full provenance: which tool, which session, when it happened.

When to Use

Search memory before doing work, not after:

Starting a new task — Check for prior decisions, context, or related work.
Making architectural decisions — Check if this was discussed before.
Debugging — Search for related past issues and solutions.
User asks about history — "what did we decide about...", "remember when...", "how did we handle..."
Encountering unfamiliar code — Check for context on why it was written that way.
Setting up or configuring something — Check for user preferences and conventions.

When in doubt, search. It costs little and prevents repeated work.

CLI Commands

Semantic Search (primary)

# Search memories by meaning
memgentic search "database migration strategy"

# Compact output — fewer tokens, good for scanning
memgentic search "database migration strategy" --format compact

# Filter by platform
memgentic search "auth implementation" -s claude_code
memgentic search "API design" -s chatgpt

# Filter by content type
memgentic search "why we chose PostgreSQL" -t decision
memgentic search "user coding style" -t preference

Store a Memory

# Save an important fact or decision
memgentic remember "We chose Qdrant over Pinecone for local-first vector storage"

# Memory type is auto-classified from content
memgentic remember "Always use UV for package management"

Other Commands

# See which platforms have memories and how many
memgentic sources

# Generate standalone context file with recent activity
memgentic update-context

# Explore knowledge graph around an entity
memgentic graph "authentication"

# Check system health
memgentic doctor

MCP Tools

If the Memgentic MCP server is running, these tools are available directly:

Tool	Purpose
`memgentic_recall`	Semantic search with source filtering
`memgentic_remember`	Store a new memory
`memgentic_search`	Full-text keyword search
`memgentic_recent`	Recent memories by timestamp
`memgentic_briefing`	Cross-agent briefing of recent activity
`memgentic_sources`	List platforms and memory counts
`memgentic_configure_session`	Set session-level source filters
`memgentic_stats`	Memory statistics and analytics
`memgentic_forget`	Archive (soft-delete) a memory
`memgentic_export`	Export memories as JSON

MCP Examples

memgentic_recall(query="authentication flow", limit=5)
memgentic_recall(query="auth", sources=["claude_code", "chatgpt"])
memgentic_remember(content="Project uses JWT with refresh tokens", memory_type="decision")
memgentic_search(query="PostgreSQL", content_type="decision")
memgentic_configure_session(exclude_sources=["codex_cli"])
memgentic_briefing(hours=24)

Efficient Retrieval Strategy

Minimize token usage with a progressive approach:

Start compact — Use --format compact for an overview of what exists.
Narrow down — Add platform filter (-s claude_code) or type filter (-t decision) when you know what you are looking for.
Full detail — Drop --format compact only when you need the complete memory content.
Limit results — Use --limit to cap the number of results returned.

# Step 1: Quick scan
memgentic search "deployment" --format compact --limit 10

# Step 2: Found relevant results, get details from a specific platform
memgentic search "deployment" -s claude_code --limit 3

Memory Types

Memories are classified into types for filtering:

Type	Contains
`decision`	Architectural and technical decisions with rationale
`learning`	Things learned during development
`preference`	User preferences, conventions, coding style
`fact`	General knowledge and project context
`bug_fix`	Bug fixes, root causes, and solutions
`conversation_summary`	Summaries of complete sessions

Filter by type to get precise results:

memgentic search "database" -t decision    # Only decisions about databases
memgentic search "testing" -t preference   # Only testing preferences
memgentic search "auth" -t bug_fix         # Only auth-related bug fixes

Cross-Platform Context

Memgentic captures from multiple AI tools. Each memory records its source platform:

Platform	Source ID
Claude Code	`claude_code`
Gemini CLI	`gemini_cli`
ChatGPT	`chatgpt`
Aider	`aider`
Codex CLI	`codex_cli`
Copilot CLI	`copilot_cli`
Claude Web/Desktop	`claude_web`
Antigravity	`antigravity`

Use source filtering to scope searches:

# What did we discuss in Claude Code about this project?
memgentic search "project architecture" -s claude_code

# What did ChatGPT suggest about this topic?
memgentic search "caching strategy" -s chatgpt

# See all available sources
memgentic sources

Capture Methods

Memories enter the system through:

auto_daemon — File watcher captures conversations automatically in the background.
mcp_tool — Stored via MCP tool calls during AI sessions.
cli — Manually saved via memgentic remember.
json_import — Bulk imported from conversation exports (ChatGPT, Claude Web).

Setup

# Check prerequisites
memgentic doctor

# Interactive setup (model selection, configuration)
memgentic setup

# Import existing conversations from all detected AI tools
memgentic import-existing

# Start background capture daemon
memgentic daemon

# Start MCP server for AI tool integration
memgentic serve

Source: Chariton-kyp/Memgentic — distributed by TomeVault.

流狐档案作者与许可取自来源；运行、权限和网络为流狐检测或估算

流狐分类: 通用
作者声明 Agent: 未找到明确声明；不据此推断已兼容或已测试
静态检查: 88 / 100 · 启发式扫描，不代表运行安全
作者 / 版本 / 许可: @tomevault-io · 未声明 license
流狐 Token 估算: 低消耗
流狐接入估算: 即装即用
是否需要外部 API Key: 未发现要求
检测到的系统要求: 未声明
底层运行要求: 未声明
检测到的文件与系统行为: 只读
允许写入 / 修改
检测到的网络行为: 仅限本地
安装命令数: 无（仅作为资料）

档案由构建时根据 SKILL.md 与安装命令自动衍生，可能与作者实际意图存在差异。

需要注意： 未限定 allowed-tools，默认拥有全部工具权限。

输出预览 memgentic.preview

# MCP Examples

memgentic_recall(query="authentication flow", limit=5)
memgentic_recall(query="auth", sources=["claude_code", "chatgpt"])
memgentic_remember(content="Project uses JWT with refresh tokens", memory_type="decision")
memgentic_search(query="PostgreSQL", content_type="decision")
memgentic_configure_session(exclude_sources=["codex_cli"])
memgentic_briefing(hours=24)

SKILL.md SKILL.md

# Memgentic — Universal AI Memory Layer

Memgentic captures knowledge from all your AI tools and makes it searchable.
Every memory carries full provenance: which tool, which session, when it happened.

## When to Use

Search memory **before** doing work, not after:

- **Starting a new task** — Check for prior decisions, context, or related work.
- **Making architectural decisions** — Check if this was discussed before.
- **Debugging** — Search for related past issues and solutions.
- **User asks about history** — "what did we decide about...", "remember when...",
  "how did we handle..."
- **Encountering unfamiliar code** — Check for context on why it was written that way.
- **Setting up or configuring something** — Check for user preferences and conventions.

When in doubt, search. It costs little and prevents repeated work.

## CLI Commands

### Semantic Search (primary)

```bash
# Search memories by meaning
memgentic search "database migration strategy"

# Compact output — fewer tokens, good for scanning
memgentic search "database migration strategy" --format compact

# Filter by platform
memgentic search "auth implementation" -s claude_code
memgentic search "API design" -s chatgpt

# Filter by content type
memgentic search "why we chose PostgreSQL" -t decision
memgentic search "user coding style" -t preference
```

### Store a Memory

```bash
# Save an important fact or decision
memgentic remember "We chose Qdrant over Pinecone for local-first vector storage"

# Memory type is auto-classified from content
memgentic remember "Always use UV for package management"
```

### Other Commands

```bash
# See which platforms have memories and how many
memgentic sources

# Generate standalone context file with recent activity
memgentic update-context

…

证据边界与执行链路

作者原文负责流程事实；流狐只索引当前章节、要点、文件与命令。

章节    -> When to Use → CLI Commands → Semantic Search (primary) → Store a Memory → Other Commands → MCP Tools
要点    -> before · Starting a new task · Making architectural decisions · Debugging · User asks about history · Encountering unfamiliar code · Setting up or configuring something · Start compact
文件/命令 -> memgenticrecall · memgenticremember · memgenticsearch · memgenticrecent · memgenticbriefing · memgenticsources · memgenticconfiguresession · memgenticstats
内容 SHA-256 -> e384a38c8a4c