| name | lwy-self |
| description | Personal alter-ego life & work log. Triggers when the user wants to record what they did today, who they collaborated with, which projects/decisions they moved forward, or says 『记一下』『今天』『周报』『我的同事/项目』『把这场会/妙记沉淀进来』『alter ego』『life log』. Files daily life and meetings into the private self store (diaries/people/products/events) so the AI becomes an informed alter-ego. Counterpart to lwy-llm-wiki (world knowledge: books/concepts); this skill owns the personal layer. |
| metadata | {"author":"learnwy","version":"1.1","privacy":"private — local-only, never pushed to the public repo"} |
lwy-self — Personal Alter-Ego
Record your life and work to build a private self-model that knows you — the personal layer,
counterpart to lwy-llm-wiki (world knowledge).
- World wiki (
~/.learnwy/llm-wiki/, public): "what I know about the world" — books, concepts, authors.
- Alter-ego / self (
~/.learnwy/ai/private/self/, private local-only submodule): "who I am, what I did, who I know" — diaries, colleagues, projects, decisions.
Storage & engine
self shares the same engine as lwy-llm-wiki (the shared module src/shared/wiki/); only the
default root differs. lwy-self now has its own binary defaulting to the private self root
(~/.learnwy/ai/private/self), so everyday commands need no --root; pass --root to point at a different store.
lwy-self stats
lwy-self lint
lwy-self generate-index
lwy-self generate-topics
lwy-self stats --root /some/other/wiki
Full operations manual lives in that root's CLAUDE.md (entity-first layout, capture rhythm, privacy conventions).
When to use this skill
| Signal | Action |
|---|
| "record this / today / I just talked with X" | Daily quick-capture: append a dated bullet to ai/private/self/wiki/diaries/<ISO-week>.md |
| "review this week / weekly report / recap the week" | Weekly review: read the week's diary, promote durable facts into people/products/events entity pages |
| "who is X / who am I doing Y with / what's the status of this project" | Query: read the self people/products/events pages and answer |
| "file this Feishu group/doc into self" | Ingest: land via lark-context into raw/lark/, then compile into threads/people/events (see lwy-llm-wiki's ingest-lark flow) |
| "file this meeting/minutes into self / record this meeting" | Meeting ingest: pull the minutes transcript → resolve speakers → compile into diaries/people/products/events (see references/ingest-meeting.md) |
Three-step workflow
- Daily quick-capture → append one dated line (event / conversation / decision) under the date in
diaries/<ISO-week>.md. One line each; land it first, lose nothing.
- Compile → when a bullet warrants persistence, fold it into the relevant
[[people/slug]] / [[products/slug]] / [[events/slug]] entity page and cross-link (compiling is integrating into the network, not isolated summarizing).
- Weekly review → read the whole week's diary, promote durable facts into entity pages, run
lint to keep links clean.
Name resolution (the key to "being there")
When filing meetings / group chats, every speaker must resolve to one specific person — get it right and
each statement attributes to a person; get it wrong and it degrades into "some colleague thought…"
summary-speak. The self store's people/ pages ARE the living name map: they already carry
alias/pinyin/homophone resolution, roles, and the relationship to you (汪洋) — stronger than a static
name-map.md, and they compound over time. When resolving, tell yourself from others and from managers
(a manager's-voice statement most likely comes from a manager); when in doubt re-check the speaker
markers and mark "(speaker TBD)" rather than mis-attributing.
Boundaries
- Privacy: PII and internal company content. Local + private submodule only; never copied into the public
llm-wiki/ world store.
- Not this skill: world knowledge (books/concepts) →
lwy-llm-wiki; generic conversation insight → lwy-knowledge-consolidation; a forwardable one-shot meeting note (terminal / Feishu card) → the meeting-notes skill (this skill only does store-ingest; the two chain).
- Chinese content is fine inside the store itself (single-user private library); skill docs stay English.
