life-capture

Turn natural-language life logs into durable records. This skill classifies each input item, generates tags, creates user-visible markdown, writes to a daily note under life/daily, and syncs structured data into life/db/life.db.

Default storage layout

Use these paths unless the user explicitly overrides them:

CODEBLOCK0

Create missing directories as needed. Never delete existing content. Append or update only.

Supported record types

Map every parsed item to exactly one primary type:

• - expense: spending, bills, purchases, subscriptions, refunds
• INLINECODE3: completed tasks, ongoing work, todos, chores, habits
• INLINECODE4: calendar items, appointments, time blocks, plans
• INLINECODE5: ideas, inspiration, possible projects, reflections worth saving

When a sentence contains multiple items, split it into multiple records.

Output contract

For each user request:

1. 1. Parse the message into one or more records.
2. Generate a stable id for each record using the pattern:

1. 3. Generate 1 to 4 short tags.
2. Show the user the organized result in markdown.
3. Save the records by running scripts/process_entry.py.

Always keep the original user wording in raw_text. Never invent missing fields. Leave unknown fields null.

User-visible response format

Because this skill is configured for visible output, show a concise but complete result after writing:

CODEBLOCK1json

CODEBLOCK2

If there are multiple records, repeat the block for each one.

Parsing rules

Use scripts/parse_entries.py for natural-language parsing. The parser now reads configurable rules from references/parser_config.json, so prefer editing that file instead of changing Python when you need new categories, tags, or keyword mappings.

Expense

Extract when present:

• - INLINECODE15
• INLINECODE16 (default CNY only when the currency symbol or language implies RMB; otherwise null)
• INLINECODE18
• INLINECODE19
• INLINECODE20
• INLINECODE21

Default top-level tags often include 开销 plus one semantic tag such as 餐饮 or 交通.

Preferred categories:

• - 饮食
• 交通
• 购物
• 居家
• 社交
• 娱乐
• 医疗
• 学习
• 其他

Task

Extract when present:

• - status (todo, doing, done, cancelled)
• INLINECODE30 (low, normal, high)
• INLINECODE34
• INLINECODE35
• INLINECODE36

If the user says they already did something, default status to done.

Schedule

Extract when present:

• - INLINECODE38
• INLINECODE39
• INLINECODE40
• INLINECODE41
• INLINECODE42 (planned, done, skipped)

If the user uses relative dates, resolve them from the current conversation date. Prefer passing --today YYYY-MM-DD to scripts/process_entry.py or scripts/parse_entries.py so relative dates like 明天 are stable across environments.

Idea

Extract when present:

• - INLINECODE50
• INLINECODE51 (captured, reviewing, used, archived)
• INLINECODE56

Default status to captured.

Configurable parsing rules

Before editing Python, check whether the change can be made in references/parser_config.json.

You can change:

• - category and subcategory mappings for expenses
• task project mappings
• idea type mappings
• schedule extra tag mappings
• default tags by record type
• hint regexes used in type inference

To test a modified config without changing the bundled default file:

CODEBLOCK3

Markdown writing rules

Write each record into the daily note for its effective date under one of these sections:

• - INLINECODE59
• INLINECODE60
• INLINECODE61
• INLINECODE62

Use this block structure:

CODEBLOCK4

Then add type-specific fields:

• - Expense: 金额 / 币种 / 分类 / 子分类 / 商家 / 支付方式
• Task: 状态 / 优先级 / 项目 / 截止日期 / 完成时间
• Schedule: 日期 / 开始时间 / 结束时间 / 地点 / 状态
• Idea: 类型 / 状态 / 关联任务

Execution workflow

End-to-end one-command flow

Use this when the user provides natural language and wants the records saved immediately:

CODEBLOCK5

The wrapper script will:

1. 1. initialize the database if missing
2. parse text into {"records": [...]} with INLINECODE64
3. save markdown and sqlite rows with INLINECODE65
4. print parsed records plus save results as json

Split-step flow

Use this when the user asks to inspect or verify the structured output before writing:

CODEBLOCK6

Then save:

CODEBLOCK7

Database init only

Use this once before first write if life/db/life.db does not exist and you are not using process_entry.py:

CODEBLOCK8

Database sync rules

The database design is:

• - INLINECODE68
• INLINECODE69
• INLINECODE70
• INLINECODE71
• INLINECODE72
• INLINECODE73
• INLINECODE74

See references/schema.md for the schema, references/examples.md for sample payloads and commands, and references/configuration.md plus references/parser_config.json for configurable parsing rules.

Failure handling

• - If markdown write succeeds but database sync fails, say so clearly.
• Do not silently drop a record.
• If parsing is ambiguous, make the narrowest safe interpretation and preserve the original text.
• If a record is missing a critical type-specific field, still save the record with null fields rather than discarding it.