wiki-agent

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Wiki Agent — Targeted Cross-Agent History Search + Ingest

Wiki Agent — 跨Agent历史定向搜索与摄入

You are doing a query-driven targeted ingest from one specific AI agent's raw conversation history. The user is typically working in a different agent right now and wants to pull in context from another agent's past sessions.

This is not bulk ingest. You find sessions about a specific topic, extract the relevant blobs, distill them into the wiki, and return a synthesized answer the user can act on immediately.

你将执行基于查询的特定AI Agent原始对话历史定向摄入。用户当前通常在另一个Agent中工作，想要从其他Agent的过往会话中调取上下文信息。

这不是批量摄入。你需要找到与特定主题相关的会话，提取相关内容片段，将其提炼到wiki中，并返回用户可立即使用的合成答案。

Command Routing

命令路由

Parse the invocation to determine the target agent and optional query:

Command	Target	Example
`/wiki-claude [query]`	Claude Code history	`/wiki-claude "how did I set up auth middleware"`
`/wiki-codex [query]`	Codex CLI history	`/wiki-codex "rust ownership patterns"`
`/wiki-hermes [query]`	Hermes agent history	`/wiki-hermes "memory architecture"`
`/wiki-openclaw [query]`	OpenClaw history	`/wiki-openclaw "project planning approach"`
`/wiki-copilot [query]`	Copilot chat history	`/wiki-copilot "test strategy for API routes"`

If no query is given, default to recent sessions mode: ingest the last 5 unprocessed sessions from that agent and return a summary of what was found. This is equivalent to a focused

wiki-history-ingest

for that agent only.

解析调用指令以确定目标Agent和可选查询：

命令	目标	示例
`/wiki-claude [query]`	Claude Code历史记录	`/wiki-claude "how did I set up auth middleware"`
`/wiki-codex [query]`	Codex CLI历史记录	`/wiki-codex "rust ownership patterns"`
`/wiki-hermes [query]`	Hermes Agent历史记录	`/wiki-hermes "memory architecture"`
`/wiki-openclaw [query]`	OpenClaw历史记录	`/wiki-openclaw "project planning approach"`
`/wiki-copilot [query]`	Copilot聊天历史记录	`/wiki-copilot "test strategy for API routes"`

如果未提供查询，默认进入近期会话模式：摄入该Agent的最后5个未处理会话，并返回找到内容的摘要。这相当于针对该Agent的聚焦版

wiki-history-ingest

。

Before You Start

开始前准备

Read

~/.obsidian-wiki/config

→ get

OBSIDIAN_VAULT_PATH

Read
```
$OBSIDIAN_VAULT_PATH/.manifest.json
```
→ know what's already ingested.
Read
```
$OBSIDIAN_VAULT_PATH/hot.md
```
if it exists → warm context on recent wiki activity.

读取

~/.obsidian-wiki/config

→ 获取

OBSIDIAN_VAULT_PATH

。

读取
```
$OBSIDIAN_VAULT_PATH/.manifest.json
```
→ 了解已摄入的内容。
如果存在
```
$OBSIDIAN_VAULT_PATH/hot.md
```
，请读取 → 获取wiki近期活动的上下文信息。

Step 1: Locate the Agent's History Root

步骤1：定位Agent的历史记录根目录

Agent	Default path	Config override
`claude`	`~/.claude` + `~/Library/Application Support/Claude/local-agent-mode-sessions/`	`CLAUDE_HISTORY_PATH` in `.env`
`codex`	`~/.codex`	`CODEX_HISTORY_PATH` in `.env`
`hermes`	`~/.hermes`	`HERMES_HOME` in env or `.env`
`openclaw`	`~/.openclaw`	`OPENCLAW_HOME` in `.env`
`copilot`	`~/.copilot`	`COPILOT_HISTORY_PATH` in `.env`

If the history root doesn't exist, stop and tell the user: "No

<agent>

history found at

<path>

. Have you run

<agent>

on this machine? You can set a custom path with

<CONFIG_VAR>

.env

Agent	默认路径	配置覆盖项
`claude`	`~/.claude` + `~/Library/Application Support/Claude/local-agent-mode-sessions/`	`.env` 中的 `CLAUDE_HISTORY_PATH`
`codex`	`~/.codex`	`.env` 中的 `CODEX_HISTORY_PATH`
`hermes`	`~/.hermes`	环境变量或 `.env` 中的 `HERMES_HOME`
`openclaw`	`~/.openclaw`	`.env` 中的 `OPENCLAW_HOME`
`copilot`	`~/.copilot`	`.env` 中的 `COPILOT_HISTORY_PATH`

如果历史记录根目录不存在，请停止操作并告知用户："在

<path>

未找到

<agent>

历史记录。你是否在此机器上运行过

<agent>

？你可以在

.env

中通过

<CONFIG_VAR>

设置自定义路径。"

Step 2: Build Session Inventory

步骤2：构建会话清单

Use the cheapest index source for each agent — don't open session files until you know which ones are relevant.

为每个Agent使用成本最低的索引源——在确定哪些会话相关之前，不要打开会话文件。

Claude

Primary index:   ~/.claude/projects/  (directories = projects, files = sessions)
Session files:   ~/.claude/projects/*/*.jsonl
Desktop index:   find ~/Library/Application Support/Claude/local-agent-mode-sessions -name "local_*.json"
Signal fields:   sessionId, cwd, startedAt, title (in local_*.json)

Build a list of sessions:

{path, project_dir, modified_at, already_ingested}

Primary index:   ~/.claude/projects/  (directories = projects, files = sessions)
Session files:   ~/.claude/projects/*/*.jsonl
Desktop index:   find ~/Library/Application Support/Claude/local-agent-mode-sessions -name "local_*.json"
Signal fields:   sessionId, cwd, startedAt, title (in local_*.json)

构建会话列表：

{path, project_dir, modified_at, already_ingested}

。

Codex

Primary index:   ~/.codex/session_index.jsonl
Session files:   ~/.codex/sessions/**/rollout-*.jsonl
Signal fields:   thread_id, name/title, updated_at (in session_index.jsonl)

Read

session_index.jsonl

as the inventory. Each line:

{thread_id, name, updated_at}

. Map thread IDs to rollout files by matching directory names.

Primary index:   ~/.codex/session_index.jsonl
Session files:   ~/.codex/sessions/**/rollout-*.jsonl
Signal fields:   thread_id, name/title, updated_at (in session_index.jsonl)

读取

session_index.jsonl

作为清单。每一行：

{thread_id, name, updated_at}

。通过匹配目录名将线程ID映射到rollout文件。

Hermes

Primary index:   ~/.hermes/memories/*.md  (fast to scan)
Session files:   ~/.hermes/sessions/**/*.jsonl
Signal fields:   file names, memory titles, first 3 lines of each memory

Scan memory filenames first (they're often titled by topic). Fall back to session listing.

Primary index:   ~/.hermes/memories/*.md  (fast to scan)
Session files:   ~/.hermes/sessions/**/*.jsonl
Signal fields:   file names, memory titles, first 3 lines of each memory

首先扫描内存文件名（通常按主题命名）。如果没有匹配项，再回退到会话列表。

OpenClaw

Primary index:   ~/.openclaw/workspace/memory/MEMORY.md  (structured long-term memory)
Daily notes:     ~/.openclaw/workspace/memory/YYYY-MM-DD.md
Session index:   ~/.openclaw/agents/*/sessions/sessions.json
Session files:   ~/.openclaw/agents/*/sessions/*.jsonl

Read

MEMORY.md

sections first — it's the pre-compiled summary of everything. Daily notes give recency signal.

Primary index:   ~/.openclaw/workspace/memory/MEMORY.md  (structured long-term memory)
Daily notes:     ~/.openclaw/workspace/memory/YYYY-MM-DD.md
Session index:   ~/.openclaw/agents/*/sessions/sessions.json
Session files:   ~/.openclaw/agents/*/sessions/*.jsonl

首先读取

MEMORY.md

的章节——这是所有内容的预编译摘要。每日笔记提供时效性信号。

Copilot

Primary index:   session filenames / directory listing
Session files:   varies by client (VS Code: ~/.copilot/sessions/*.jsonl or similar)
Signal fields:   session timestamps, file names

Primary index:   session filenames / directory listing
Session files:   varies by client (VS Code: ~/.copilot/sessions/*.jsonl or similar)
Signal fields:   session timestamps, file names

Step 3: Score Sessions Against the Query

步骤3：根据查询为会话评分

If a query was given, score each session in the inventory without opening full session files:

Name/title match — does the session name or thread title contain the query terms? Score: +3
CWD/project match — does the working directory suggest the right project? Score: +2
Recency — sessions from the last 90 days score higher than older ones. Score: +1 per 30-day recency bracket (max +3)
Already ingested — if this session was previously ingested and the wiki page already covers the query (check
```
hot.md
```
+
```
index.md
```
), flag as "covered" but still show in results

Select the top 3–5 sessions by score. If no query was given, select the 5 most recent unprocessed sessions.

如果提供了查询，在不打开完整会话文件的情况下为清单中的每个会话评分：

名称/标题匹配——会话名称或线程标题是否包含查询词？评分：+3
工作目录/项目匹配——工作目录是否指向正确的项目？评分：+2
时效性——近90天内的会话比旧会话得分更高。评分：每30天时间段+1（最高+3）
已摄入状态——如果该会话之前已被摄入，且wiki页面已覆盖该查询（检查
```
hot.md
```
+
```
index.md
```
），标记为“已覆盖”但仍显示在结果中

选择得分前3-5的会话。如果未提供查询，选择5个最新的未处理会话。

Step 4: Extract the Relevant Blob

步骤4：提取相关内容片段

Open each selected session file and extract only the content relevant to the query. Do not read the full session if it's large — use targeted extraction.

打开每个选中的会话文件，仅提取与查询相关的内容。如果会话文件较大，不要读取完整内容——使用定向提取。

Per-Agent Extraction Strategy

各Agent提取策略

Claude (JSONL conversation):

Each line:
```
{role, content, timestamp, ...}
```
Search with:
```
grep -i "<query terms>" <session.jsonl>
```
to find the relevant lines
Extract: the surrounding conversation window (10 lines before + 20 lines after each hit)
Special signal: tool calls (Read/Write/Bash/Edit) reveal what was actually done — extract these even without keyword matches if they're in the relevant window

Codex (rollout JSONL):

Each line:

{type: "session_meta|turn_context|event_msg|response_item", ...}

Filter to
```
type: "event_msg"
```
(user turns) and
```
type: "response_item"
```
(model output)
Search with:
```
grep -i "<query terms>" <rollout.jsonl>
```
Extract: matching turns + their parent context (the
```
turn_context
```
preceding the match)
Skip:
```
session_meta
```
events (operational metadata, not knowledge)

Hermes (memory files + session JSONL):

For memory files: read the full file (they're short — typically <500 words each)
For session JSONL:
```
grep -i "<query terms>"
```
+ surrounding window
Memory files with title matches → read fully; others → grep only

OpenClaw (MEMORY.md + daily notes + session JSONL):

```
MEMORY.md
```
: grep for section headers containing query terms → extract that section
Daily notes: grep most recent 30 days for query terms → extract matching paragraphs
Session JSONL: same grep-window approach as Claude
Prefer MEMORY.md/daily notes over session JSONL (they're pre-synthesized)

Copilot (session JSONL):

Same grep-window approach as Claude
Look for checkpoint files if available (pre-summarized)

Claude（JSONL对话）：

每一行：
```
{role, content, timestamp, ...}
```
搜索方式：
```
grep -i "<query terms>" <session.jsonl>
```
查找相关行
提取内容：匹配内容前后的对话窗口（匹配项前10行+后20行）
特殊信号：工具调用（Read/Write/Bash/Edit）显示实际操作内容——即使没有关键词匹配，只要在相关窗口内也要提取这些内容

Codex（rollout JSONL）：

每一行：

{type: "session_meta|turn_context|event_msg|response_item", ...}

筛选
```
type: "event_msg"
```
（用户轮次）和
```
type: "response_item"
```
（模型输出）
搜索方式：
```
grep -i "<query terms>" <rollout.jsonl>
```
提取内容：匹配的轮次及其父上下文（匹配项之前的
```
turn_context
```
）
跳过：
```
session_meta
```
事件（操作元数据，非知识内容）

Hermes（内存文件+会话JSONL）：

对于内存文件：读取完整文件（文件较短，通常<500词）
对于会话JSONL：使用
```
grep -i "<query terms>"
```
+ 上下文窗口
标题匹配的内存文件→完整读取；其他→仅使用grep搜索

OpenClaw（MEMORY.md+每日笔记+会话JSONL）：

```
MEMORY.md
```
：搜索包含查询词的章节标题→提取该章节
每日笔记：搜索最近30天内的查询词→提取匹配段落
会话JSONL：采用与Claude相同的grep窗口方法
优先使用MEMORY.md/每日笔记而非会话JSONL（它们是预合成的）

Copilot（会话JSONL）：

采用与Claude相同的grep窗口方法
如果有 checkpoint 文件（预摘要），优先使用

Step 5: Distill Blobs into Wiki Pages

步骤5：将内容片段提炼到Wiki页面

For each extracted blob, determine where it belongs in the wiki:

Check if a wiki page already covers this — grep
```
index.md
```
and page frontmatter for the topic. If yes, update the existing page rather than creating a new one.
Determine category using standard rules (from
```
llm-wiki/SKILL.md
```
):
- Technique / how-to →
```
skills/
```
- Abstract concept / pattern →
```
concepts/
```
- Tool / library / person →
```
entities/
```
- Cross-cutting insight →
```
synthesis/
```

Write or update the page with required frontmatter:

yaml

---
title: <topic>
category: skill|concept|entity|synthesis
tags: [tag1, tag2]
sources: [<agent>://<path/to/session>]
created: <date>
updated: <date>
confidence: high|medium|low
lifecycle: stable|draft
---

Set

sources

with the agent prefix so

memory-bridge

can find it later.

Add cross-links to related wiki pages found in
```
index.md
```
.

Distillation rules (same as all ingest skills):

Extract durable knowledge, not operational telemetry
One wiki page per concept, not one per session
Merge into existing pages rather than duplicating
Keep the signal: decisions made, patterns discovered, techniques that worked, bugs explained

对于每个提取的内容片段，确定其在wiki中的归属：

检查是否已有wiki页面覆盖该主题——在
```
index.md
```
和页面前置元数据中搜索该主题。如果有，更新现有页面而非创建新页面。
确定分类使用标准规则（来自
```
llm-wiki/SKILL.md
```
）：
- 技术/操作指南 →
```
skills/
```
- 抽象概念/模式 →
```
concepts/
```
- 工具/库/人物 →
```
entities/
```
- 跨领域见解 →
```
synthesis/
```

编写或更新页面并添加所需前置元数据：

yaml

---
title: <topic>
category: skill|concept|entity|synthesis
tags: [tag1, tag2]
sources: [<agent>://<path/to/session>]
created: <date>
updated: <date>
confidence: high|medium|low
lifecycle: stable|draft
---

设置

sources

时添加Agent前缀，以便

memory-bridge

后续能找到它。

添加交叉链接到
```
index.md
```
中找到的相关wiki页面。

提炼规则（与所有摄入技能相同）：

提取持久知识，而非操作遥测数据
每个概念对应一个wiki页面，而非每个会话对应一个
合并到现有页面而非重复创建
保留关键信息：做出的决策、发现的模式、有效的技术、问题解释

Step 6: Return Synthesized Answer

步骤6：返回合成答案

After ingesting, immediately synthesize and return an answer from the newly ingested + existing wiki content:

undefined

摄入完成后，立即从新摄入的内容+现有wiki内容中合成并返回答案：

undefined

From <agent> history: "<query>"

来自<agent>历史记录："<query>"

Found in: <N> sessions (<session names/titles>)

Key insights: <Synthesized answer — 3–5 bullet points of the most useful knowledge>

Wiki pages updated/created:

[[page-name]] — <what was added>
[[page-name]] — <what was added>

Sessions ingested:

Session	Date	Relevance
<name>	<date>	<one-line why it was selected>

Gaps: <What the sessions didn't cover that might be relevant>


If a query was given but no relevant sessions were found, say so explicitly: "No sessions about '<query>' found in `<agent>` history. The most recent sessions covered: <list topics from last 3 sessions>."

---

找到于： <N>个会话（<会话名称/标题>）

关键见解： <合成答案——3-5个最有用知识的要点>

更新/创建的Wiki页面：

[[page-name]] — <添加的内容>
[[page-name]] — <添加的内容>

已摄入的会话：

会话	日期	相关性
<name>	<date>	<选中原因的单行说明>

空白点： <会话未涵盖但可能相关的内容>


如果提供了查询但未找到相关会话，请明确告知："在`<agent>`历史记录中未找到关于'<query>'的会话。最近的会话涵盖：<最近3个会话的主题列表>。"

---

Step 7: Update Tracking Files

步骤7：更新跟踪文件

Update

.manifest.json

for each session file processed:

json

{
  "<path>": {
    "ingested_at": "<now>",
    "source_type": "<agent>_conversation",
    "modified_at": "<file mtime>",
    "pages_created": [...],
    "pages_updated": [...]
  }
}

Append to

log.md

- [TIMESTAMP] WIKI-AGENT agent=<agent> query="<query>" sessions_searched=N sessions_ingested=M pages_created=X pages_updated=Y

Update

hot.md

with a one-line summary of what was ingested.

为每个处理的会话文件更新

.manifest.json

：

json

{
  "<path>": {
    "ingested_at": "<now>",
    "source_type": "<agent>_conversation",
    "modified_at": "<file mtime>",
    "pages_created": [...],
    "pages_updated": [...]
  }
}

向

log.md

追加内容：

- [TIMESTAMP] WIKI-AGENT agent=<agent> query="<query>" sessions_searched=N sessions_ingested=M pages_created=X pages_updated=Y

用摄入内容的单行摘要更新

hot.md

。

Cross-Agent Use Patterns

跨Agent使用模式

These are the primary use cases this skill is designed for:

"I'm on Codex. What did I figure out about X in Claude?" →

/wiki-claude "X"

— finds Claude sessions about X, ingests them, returns the answer

"I solved a bug in Hermes last week. I need that context now in Claude Code." →

/wiki-hermes "bug description"

— surfaces and ingests the Hermes session

"What are all the approaches I've tried for X across all my tools?" → Run

/wiki-claude "X"

/wiki-codex "X"

/wiki-hermes "X"

in sequence — each ingests its slice, the wiki accumulates the cross-agent picture, then

/memory-bridge diff

shows what each tool uniquely contributed

No query — just "catch me up on recent Codex work" →

/wiki-codex

— ingests last 5 Codex sessions and returns a summary

以下是该技能设计的主要使用场景：

"我正在使用Codex。我在Claude中关于X的解决方案是什么？" →

/wiki-claude "X"

— 查找Claude中关于X的会话，摄入并返回答案

"我上周在Hermes中解决了一个bug。现在需要在Claude Code中使用该上下文。" →

/wiki-hermes "bug description"

— 调取并摄入Hermes会话

"我在所有工具中尝试过哪些解决X的方法？" → 依次运行

/wiki-claude "X"

、

/wiki-codex "X"

、

/wiki-hermes "X"

— 每个工具摄入其对应的内容片段，wiki积累跨Agent的完整信息，然后

/memory-bridge diff

展示每个工具的独特贡献

无查询——仅需"了解我最近在Codex中的工作内容" →

/wiki-codex

— 摄入最后5个Codex会话并返回摘要