nt-learn
/nt-learn
Zbiera trwałe wnioski z sesji do magazynu nt: lekcje, reguły, notatki i zadania, każdą pozycję proponując do zatwierdzenia.
- Autor
- navbytes ↗
- Repozytorium
- nt ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 21.09.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 21.09.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Zbiera trwałe wnioski z sesji do magazynu nt: lekcje, reguły, notatki i zadania, każdą pozycję proponując do zatwierdzenia.
Wyzwalacze z opisu
Oryginalny opis (EN)
Harvest this session's durable learnings into the user's nt store — lessons, rules, core memory, notes, and follow-up tasks. Use when the user asks to save what was learned, record lessons or mistakes from this session, remember something for future sessions, or when they type /nt-learn. Proposes every item and waits for approval; never writes silently.
Jak działa
- 1. Extract candidates
- 2. Dedup before proposing
- 3. Present for approval — do not write yet
- 4. Write the approved items
Podgląd SKILL.md
---
name: nt-learn
description: Harvest this session's durable learnings into the user's nt store — lessons, rules, core memory, notes, and follow-up tasks. Use when the user asks to save what was learned, record lessons or mistakes from this session, remember something for future sessions, or when they type /nt-learn. Proposes every item and waits for approval; never writes silently.
---
Review this session and harvest what should outlive it into nt — the durable
memory store. Optional focus: "$ARGUMENTS" (if non-empty, only propose items
related to it).
## 1. Extract candidates
Go back over the whole conversation and collect candidates in these buckets:
- **lesson** — a mistake/footgun/dead-end actually hit (a failed approach, a
wrong assumption, a command that broke something).
- **rule** — a stable directive the user stated or corrected you toward
("always X", "never Y", style/process).
- **memory-core** — a small durable fact about the user or project (preferences,
conventions, environment quirks) every future session should know.
- **note** — a decision, finding, or constraint worth reading back later.
- **task** — concrete unfinished follow-up work.
Then **skip** anything re-discoverable by rereading the code/docs, one-off state,
or raw data/log dumps — memory is for what you *couldn't* re-derive, not a
transcript.
Be **stingy** with `rule` and `memory-core` — they're injected into every future
session and cost tokens forever; propose them only on clear evidence they're
durable and always-relevant. Be **generous** with lessons and notes — free until
retrieved. If the session produced nothing worth saving, say so and stop; don't
invent learnings.
## 2. Dedup before proposing
For each candidate, check it isn't already recorded (`nt_recall` the gist,
`nt_search` its key terms). Drop duplicates. If an existing note covers the same
ground but is outdated, propose an **update**/**supersede** of it (by id) instead.
## 3. Present for approval — do not write yet
Show a numbered list. Per item: bucket, proposed **title**, one-line
**description** (lessons in trigger-form "when X, do Y — not Z"), a
**`--project` or topical `--tag`** (never omit it — the structural
`lesson`/`rule` tag is stripped before the duplicate check, so a note carrying
only that one silently bypasses the guard and duplicates an existing lesson),
and target folder+tag (`lessons/`+`lesson`, `rules/`+`rule`,
`memory/`+`memory-core`, `decisions/`/`ref/` for notes). Flag `rule`/`memory-core` items ⚠
*always-in-context — costs tokens on every future request*. (The reverse pass —
pruning rules that stopped earning that cost — is `/nt-distill rules`.) Then ask which
to save (all, numbers, edits, or none) and **wait**. Save nothing without
approval.
## 4. Write the approved items
- Notes/rules/memory/lessons → `nt_note` (title, `description`, `body`;
`kind:"lesson"|"rule"|"decision"|"ref"|"memory"` applies the canonical
tag+folder). Set `source:"claude"`, and pass **`if_exists:"return"`** — on
`matched:true` the topic's canonical note already exists: update it via
`nt_note_edit` (with the returned `expect_mtime`), and if the update reverses
a conclusion, record why with `nt_decide`. Give volatile facts a
`half_life` (e.g. `"90d"`) so they fade instead of going stale. Apply the
user's edits.
- Tasks → `nt_add` (verb-first title ≤ ~60 chars, detail in `body`, link via
`discovered_from`).
- If `nt_note`'s response has a `similar` list, check whether you doubled a note;
if so, consolidate — extend the original (`nt_note_edit` append or
`old_string`+`new_string`; CLI `nt edit --append`) or retire the duplicate
(`nt_archive superseded_by=<kept id>`).
- Finish with a short receipt: each saved item's id and where it went.
If the `nt_*` tools are unavailable, fall back to the `nt` CLI over bash (`nt
note …`, `nt add …`, `nt recall …`; multi-line/backtick bodies via `--body-file
-`); if that's missing too, tell the user to run `nt mcp install` and stop.
Wersje
W bazie nie ma innych skilli o nazwie nt-learn.
Migawka bazy z 13.11.2026 · wersja paczki eef136a5a696 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100learn
Zapisuje wnioski i decyzje z sesji do pliku .suiperpower/learnings.md, np. przy podsumowaniu pracy lub rejestrowaniu decyzji. - 100learn
Wyciąga najważniejsze wnioski z bieżącej rozmowy i zapisuje je w pliku docs/learnings.md. - 100nt-distill
Porządkuje magazyn nt: scala podobne notatki i odchudza zasady ładowane do każdej sesji, po zatwierdzeniu. - 100reflect
Wydobywa trwałe wnioski z sesji i wskazuje dla każdego miejsce zapisu: pamięć, notatki projektu, instrukcje lub zgłoszenie narzędziowe. - 90memory-discipline
Zapisuje wnioski z projektu w plikach pamięci w formacie lessons i MEMORY.md, według zasad czytania i pisania wpływających na długie zadania. - 90reflect
Zbiera z sesji wnioski warte podzielenia się i publikuje je w Crew.