harvest

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Harvest

Capture high-signal conversation knowledge into

docs/notes/

for reuse across sessions.

Terminology:
harvest
is the skill name (the action of capturing knowledge). The second brain it builds belongs to the project — name it
[Project Name] Second Brain
, not "Harvest".

将高价值对话知识捕获到

docs/notes/

目录中，以便在不同会话间复用。

术语说明：
harvest
是技能名称（即捕获知识的操作）。它构建的第二大脑属于项目，命名为
[Project Name] Second Brain
，而非“Harvest”。

When to Trigger

触发时机

Explicit request:

"harvest", "/harvest"
"harvest this", "harvest this conversation"
"save this to second brain"
"document this work", "capture this knowledge"

Natural breakpoints — AI may suggest (never auto-execute):

Significant decision made
Complex problem solved after multiple attempts
Important lesson learned
Discussion wrapping up with valuable content

We just [made a decision/solved X/learned Y].
Would you like me to update the second brain?

Wait for user confirmation. Never auto-execute.

明确请求：

"harvest"、"/harvest"
"harvest this"、"harvest this conversation"
"save this to second brain"
"document this work"、"capture this knowledge"

自然断点——AI可建议（绝不能自动执行）：

做出重大决策时
多次尝试后解决复杂问题时
学到重要经验时
讨论结束且内容有价值时

我们刚刚[做出了一项决策/解决了问题X/学到了经验Y]。
是否需要我更新第二大脑？

等待用户确认，绝不能自动执行。

Non-Negotiables

不可违背规则

Never auto-execute. Always wait for explicit user confirmation.
Use template files as the single source of structure.
Keep
```
context_id
```
in frontmatter for smart merge.
Keep MOCs as index files: link to context anchors, do not duplicate full lesson content.
Omit optional empty sections. Do not write empty headers or "None" placeholders.
Keep behavior logic in this file; keep structure/manifest data in
```
references/
```
.
Use list-first formatting for context content. Use tables only for short, fixed-width, easy-to-scan fields.

绝不能自动执行，始终等待用户明确确认。
将模板文件作为结构的唯一来源。
在前置元数据中保留
```
context_id
```
以实现智能合并。
将MOC作为索引文件：链接到上下文锚点，不要复制完整的经验内容。
省略可选的空章节，不要写入空标题或“无”占位符。
行为逻辑保留在本文件中；结构/清单数据存放在
```
references/
```
目录下。
上下文内容优先使用列表格式。仅在字段简短、固定宽度且易于扫描时使用表格。

Quick Start

快速开始

Ensure
```
docs/notes/
```
exists (initialize when missing).
Run lesson review for this harvest run.
Detect
```
context_id
```
and check whether matching context file exists.
Create new context or smart-merge existing context.
Update
```
docs/notes/00-INDEX.md
```
and related MOCs.
Confirm created/updated files to user.

确保
```
docs/notes/
```
目录存在（若缺失则初始化）。
针对本次Harvest运行进行经验回顾。
检测
```
context_id
```
并检查是否存在匹配的上下文文件。
创建新上下文或智能合并现有上下文。
更新
```
docs/notes/00-INDEX.md
```
及相关MOC文件。
向用户确认已创建/更新的文件。

Workflow

工作流程

Phase 1: Prepare

阶段1：准备

Initialize second brain storage when
```
docs/notes/
```
is missing:
- Use references/initialization-manifest.md as the single source for initialization inventory.
- Follow idempotent behavior: create missing files/directories, preserve existing files unless explicit update is requested.
- Check
```
AGENTS.md
```
  then
```
CLAUDE.md
```
  ; if both exist or neither exists, ask user which file to append, then append section from references/agents-lessons-section.md.
- Do not own or maintain
```
.base
```
  template syntax in this skill. If
```
obsidian-bases
```
  is available, delegate default base-file creation to
```
obsidian-bases
```
  using references/bases-generation-spec.md.
Mandatory lesson review for this harvest run:
- Read
```
docs/notes/00-INDEX.md
```
  when available.
- Scan
```
docs/notes/mocs/lessons-learned.md
```
  when available.
- Apply matched lessons (tech, operation type, error pattern).

当
```
docs/notes/
```
目录缺失时初始化第二大脑存储：
- 以[references/initialization-manifest.md]作为初始化清单的唯一来源。
- 遵循幂等行为：创建缺失的文件/目录，除非明确要求更新，否则保留现有文件。
- 检查
```
AGENTS.md
```
  和
```
CLAUDE.md
```
  ；若两者都存在或都不存在，询问用户要追加到哪个文件，然后从[references/agents-lessons-section.md]追加内容。
- 本技能不负责维护
```
.base
```
  模板语法。若
```
obsidian-bases
```
  可用，将默认基础文件的生成委托给
```
obsidian-bases
```
  ，使用[references/bases-generation-spec.md]。
针对本次Harvest运行进行强制经验回顾：
- 若
```
docs/notes/00-INDEX.md
```
  存在则阅读该文件。
- 若
```
docs/notes/mocs/lessons-learned.md
```
  存在则扫描该文件。
- 应用匹配的经验（技术、操作类型、错误模式）。

Phase 2: Detect + Route

阶段2：检测与路由

Build
```
context_id
```
:
- Prefer helper script
```
scripts/context_id.py
```
  (Python stdlib only, no third-party packages).
- Default call:
```
python3 scripts/context_id.py --format json
```
  (or
```
python
```
  when
```
python3
```
  is unavailable).
- Stateless policy (no state file): env keys (
```
OPENCODE_*_ID
```
  /
```
*_ID
```
  ) -> optional
```
--infer-latest-session
```
  -> generated
```
ctx-YYYYMMDDHHMMSS-<hex6>
```
  .
- Do not pass
```
--infer-latest-session
```
  by default. Use it only when the user explicitly requests latest-session inference.
- If script is unavailable or fails, use inline fallback: first env var matching
```
*SESSION*ID*
```
  /
```
*CONVERSATION*ID*
```
  /
```
*THREAD*ID*
```
  ; otherwise generate
```
ctx-YYYYMMDDHHMMSS-<hex6>
```
  .
Look for
```
docs/notes/contexts/<context_id>-*.md
```
.

Condition	Action
Matching context file exists	Run Phase 4 (Smart Merge)
No matching context file	Run Phase 3 (New Context)
User rejects suggested filename	Ask for new slug and regenerate filename
User cancels confirmation	Stop without writing files

生成
```
context_id
```
：
- 优先使用辅助脚本
```
scripts/context_id.py
```
  （仅使用Python标准库，无第三方依赖）。
- 默认调用方式：
```
python3 scripts/context_id.py --format json
```
  （若
```
python3
```
  不可用则使用
```
python
```
  ）。
- 无状态策略（无状态文件）：环境变量（
```
OPENCODE_*_ID
```
  /
```
*_ID
```
  ）→ 可选
```
--infer-latest-session
```
  → 生成
```
ctx-YYYYMMDDHHMMSS-<hex6>
```
  。
- 默认不传递
```
--infer-latest-session
```
  参数，仅当用户明确请求最新会话推断时使用。
- 若脚本不可用或执行失败，使用内联回退方案：优先匹配
```
*SESSION*ID*
```
  /
```
*CONVERSATION*ID*
```
  /
```
*THREAD*ID*
```
  的环境变量；否则生成
```
ctx-YYYYMMDDHHMMSS-<hex6>
```
  。
查找
```
docs/notes/contexts/<context_id>-*.md
```
文件。

条件	操作
存在匹配的上下文文件	执行阶段4（智能合并）
无匹配的上下文文件	执行阶段3（新上下文）
用户拒绝建议的文件名	询问新的slug并重新生成文件名
用户取消确认	停止操作，不写入任何文件

Phase 3: New Context

阶段3：新上下文

Extract high-signal items: work, decisions (with rationale), unsolved, lessons, optional source notes.
- Use stable item IDs for merge safety:
```
D-*
```
  ,
```
Q-*
```
  ,
```
LL-*
```
  .
- Keep technical signal high: merge routine process/SOP chatter into one short note instead of many standalone decisions.
- If planning files (
```
task_plan.md
```
  ,
```
findings.md
```
  ,
```
progress.md
```
  ) are present, capture medium-density snapshots (
```
conclusion + evidence + source note
```
  ) into context content.
- Do not copy full planning files. Do not create Obsidian wikilinks to files outside
```
docs/notes/
```
  .
- Exclude harvest-process artifacts from captured knowledge (command menus, filename suggestion prompts, merge/update status lines, and other capture bookkeeping).
Generate filename
```
<context_id>-<topic-slug>.md
```
and confirm:

Found: [N] decisions, [N] unsolved, [N] lessons
Suggested: contexts/<context_id>-<topic-slug>.md
1. Use this  2. Change slug  3. Cancel

Create
```
docs/notes/contexts/<filename>.md
```
from references/context-template.md.
Use
```
obsidian-markdown
```
when available for wikilinks/frontmatter/anchors.
Update
```
docs/notes/00-INDEX.md
```
according to references/index-template.md:
- Recent Updates (top 5), Topics, Key Decisions, Open Questions, Recent Lessons, Stats.
Manage MOCs:
- Lessons MOC: for error-related lessons (>15 min impact), ensure
```
docs/notes/mocs/lessons-learned.md
```
  exists via references/lessons-learned-moc-template.md, then append links only.
- Topic MOC: when a topic appears in 3+ contexts without MOC, ask user first; create from references/moc-template.md after confirmation.
Confirm:

✓ Created: contexts/<filename>.md
✓ Updated: 00-INDEX.md
✓ Updated: mocs/lessons-learned.md (when relevant)
✓ Created: mocs/<topic>.md (when relevant)

提取高价值内容：工作内容、决策（含理由）、未解决问题、经验、可选的源笔记。
- 使用稳定的条目ID确保合并安全：
```
D-*
```
  、
```
Q-*
```
  、
```
LL-*
```
  。
- 保持技术价值密度：将常规流程/SOP对话合并为一条简短笔记，而非多个独立决策。
- 若存在规划文件（
```
task_plan.md
```
  、
```
findings.md
```
  、
```
progress.md
```
  ），将中等密度快照（
```
结论+证据+源笔记
```
  ）捕获到上下文内容中。
- 不要复制完整的规划文件。不要创建指向
```
docs/notes/
```
  目录外文件的Obsidian维基链接。
- 排除Harvest流程相关的 artifacts（命令菜单、文件名建议提示、合并/更新状态行及其他捕获操作记录）。
生成文件名
```
<context_id>-<topic-slug>.md
```
并确认：

已提取：[N]项决策，[N]个未解决问题，[N]条经验
建议文件名：contexts/<context_id>-<topic-slug>.md
1. 使用此文件名  2. 修改slug  3. 取消

基于[references/context-template.md]创建
```
docs/notes/contexts/<filename>.md
```
文件。
若
```
obsidian-markdown
```
可用，使用其生成维基链接/前置元数据/锚点。
根据[references/index-template.md]更新
```
docs/notes/00-INDEX.md
```
：
- 最近更新（前5条）、主题、关键决策、未解决问题、近期经验、统计数据。
管理MOC文件：
- 经验MOC：对于影响时间超过15分钟的错误相关经验，确保
```
docs/notes/mocs/lessons-learned.md
```
  通过[references/lessons-learned-moc-template.md]存在，然后仅追加链接。
- 主题MOC：当某个主题出现在3个以上上下文但无对应MOC时，先询问用户；确认后基于[references/moc-template.md]创建。
确认完成：

✓ 已创建：contexts/<filename>.md
✓ 已更新：00-INDEX.md
✓ 已更新：mocs/lessons-learned.md（若相关）
✓ 已创建：mocs/<topic>.md（若相关）

Phase 4: Smart Merge

阶段4：智能合并

Read the existing context file.
Merge new items from current conversation.

Section	Existing Topic	New Item
Decisions Made	Match by `D-*` first, update entry and preserve anchor	Append
Still Unsolved	Match by `Q-*` ; move resolved items to Decisions with trace	Append
Lessons Learned	Match by `LL-*` first; merge same lesson and preserve anchor	Append
What We Worked On	Keep existing	Append
Source Notes	Merge by source note signature if duplicated	Append

Update frontmatter (
```
updated
```
, tags), keep
```
created
```
unchanged.
Update
```
00-INDEX.md
```
stats and recent updates.
Update
```
mocs/lessons-learned.md
```
with links only when new error-related lessons were added.
Confirm:

✓ Updated: contexts/<filename>.md
Changes: [added/updated/moved items]

读取现有上下文文件。
合并当前对话中的新条目。

章节	现有主题	新条目
已做出的决策	优先按 `D-*` 匹配，更新条目并保留锚点	追加
未解决问题	按 `Q-*` 匹配；将已解决的条目移动到决策章节并保留追踪记录	追加
学到的经验	优先按 `LL-*` 匹配；合并相同经验并保留锚点	追加
我们的工作内容	保留现有内容	追加
源笔记	按源笔记签名合并重复内容	追加

更新前置元数据（
```
updated
```
、标签），保留
```
created
```
字段不变。
更新
```
00-INDEX.md
```
中的统计数据和最近更新。
若添加了新的错误相关经验，仅更新
```
mocs/lessons-learned.md
```
中的链接。
确认完成：

✓ 已更新：contexts/<filename>.md
变更内容：[添加/更新/移动的条目]

Content Quality Rules

内容质量规则

Quality over quantity: concise, reusable, high-signal notes.
One idea per bullet; include rationale for decisions.
Keep content relevant for future reuse (3+ months horizon).
Skip raw transcripts, dead ends without insight, and obvious process noise.
Exclude harvest operational chatter: confirmation menus,
```
created/updated/skipped
```
summaries, and "how harvest processed this" narration unless it records a reusable lesson.
Use stable IDs (
```
D-*
```
,
```
Q-*
```
,
```
LL-*
```
) with anchored headings for merge-safe entries.
Prefer list-first sections for long text. Use tables only when each cell stays short (for example
```
ID | Question | Next
```
).
Avoid duplicate restatement: decision headings should carry title intent; item body should add why/impact.
For lessons, default to compact triad (
```
Issue
```
,
```
Root Cause
```
,
```
Fix
```
); add
```
Guardrail
```
/
```
Apply When
```
only when they provide non-obvious value.
Avoid repeating
```
Related
```
links on every item; prefer one section-level related MOC link.
Omit low-signal metadata (
```
Deadline
```
,
```
Carry-Over
```
) unless it materially changes follow-up decisions.
For carry-over items from prior sessions, label clearly and avoid presenting them as new outcomes.
For planning-derived content, store medium-density snapshots (
```
conclusion + evidence + source note
```
) in context files; avoid external-file wikilinks outside
```
docs/notes/
```
.

Recommended section limits:

Section	Requirement	Typical Size
Summary	Required	1-2 sentences
What We Worked On	Required	5-7 bullets
Decisions	Optional	up to 5 items
Still Unsolved	Optional	up to 3-5 items
Lessons Learned	Optional	up to 3 items
Source Notes	Optional	up to 3-5 items
Notes	Optional	short snippets only

质量优先：简洁、可复用、高价值的笔记。
每个项目符号对应一个想法；决策需包含理由。
内容需对未来复用有价值（以3个月以上为视角）。
跳过原始转录、无洞见的死胡同和明显的流程噪音。
排除Harvest操作相关的对话：确认菜单、
```
已创建/已更新/已跳过
```
摘要、以及“Harvest如何处理此内容”的说明，除非这些内容记录了可复用的经验。
使用稳定ID（
```
D-*
```
、
```
Q-*
```
、
```
LL-*
```
）和锚定标题确保条目可安全合并。
长文本章节优先使用列表格式。仅当每个单元格内容简短时使用表格（例如
```
ID | 问题 | 下一步
```
）。
避免重复表述：决策标题应明确意图；条目正文补充原因/影响。
经验内容默认使用紧凑三元组（
```
问题
```
、
```
根本原因
```
、
```
解决方案
```
）；仅当
```
防护措施
```
/
```
适用场景
```
提供非显而易见的价值时才添加。
避免在每个条目上重复
```
相关
```
链接；优先使用章节级别的相关MOC链接。
省略低价值元数据（
```
截止日期
```
、
```
待跟进
```
），除非这些内容会实质性影响后续决策。
对于来自之前会话的待跟进条目，需明确标注，避免呈现为新成果。
对于来自规划文件的内容，在上下文文件中存储中等密度快照（
```
结论+证据+源笔记
```
）；避免创建指向
```
docs/notes/
```
目录外文件的外部维基链接。

推荐章节限制：

章节	要求	典型规模
摘要	必填	1-2句话
我们的工作内容	必填	5-7个项目符号
决策	可选	最多5项
未解决问题	可选	最多3-5项
学到的经验	可选	最多3项
源笔记	可选	最多3-5项
备注	可选	仅简短片段

Integration

集成

Recommended:

obsidian-markdown

skill for enhanced Obsidian compatibility. Without it, AI may write files directly and suggest installation:

For optimal Obsidian compatibility, consider installing obsidian-markdown skill:
npx skills add <obsidian-markdown-repo>

Optional companion:

obsidian-bases

for

.base

file generation.

This skill does not own
```
.base
```
syntax templates.
When
```
obsidian-bases
```
is available, delegate generation of default base files to that skill.
Keep desired base content in references/bases-generation-spec.md, not
```
.base
```
syntax templates.
If
```
obsidian-bases
```
is unavailable, skip
```
.base
```
creation and continue markdown-only harvest flow.

Optional companion:

planning-with-files

for stronger in-progress capture.

This skill remains fully functional without it.
When available, this skill may consume planning outputs (
```
task_plan.md
```
,
```
findings.md
```
,
```
progress.md
```
) as snapshot sources, then persist reusable knowledge into
```
docs/notes/contexts/
```
.
Prefer provenance text over external links when planning files are outside Obsidian vault scope.

另请参阅

Use these files as references (single source for structure and formats).

context-template.md
moc-template.md
index-template.md
lessons-learned-moc-template.md
bases-generation-spec.md
agents-lessons-section.md
initialization-manifest.md
context_id.py

以下文件作为参考（结构和格式的唯一来源）。

context-template.md
moc-template.md
index-template.md
lessons-learned-moc-template.md
bases-generation-spec.md
agents-lessons-section.md
initialization-manifest.md
context_id.py

harvest

Original

Translation

Harvest

Harvest

When to Trigger

触发时机

Non-Negotiables

不可违背规则

Quick Start

快速开始

Workflow

工作流程

Phase 1: Prepare

阶段1：准备

Phase 2: Detect + Route

阶段2：检测与路由

Phase 3: New Context

阶段3：新上下文

Phase 4: Smart Merge

阶段4：智能合并

Content Quality Rules

内容质量规则

Integration

集成

See Also

另请参阅