fluent-journal-validation
/fluent-journal-validation
Ocenia dziennik Fluent offline względem indeksu poleceń, sprawdza kolejność i zakończenia wierszy i wydaje werdykt: czysty, ostrzeżenie albo zablokowany.
- Autor
- AstDen ↗
- Repozytorium
- kady-fluent-skills ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 12.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 12.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Ocenia dziennik Fluent offline względem indeksu poleceń, sprawdza kolejność i zakończenia wierszy i wydaje werdykt: czysty, ostrzeżenie albo zablokowany.
Oryginalny opis (EN)
Judges a Fluent journal offline against the vendored TUI index, classifies every line, checks order and terminators, and returns a clean, warn or blocked verdict. Use when a run is being considered, after a hand edit, or when the user asks to check a journal. Not for building journals, choosing physics, or running Fluent.
Jak działa
- When to use / when not to
- Inputs
- Procedure
- Outputs
- Refusals
- Hand-off
Podgląd SKILL.md
---
name: fluent-journal-validation
description: Judges a Fluent journal offline against the vendored TUI index, classifies every line, checks order and terminators, and returns a clean, warn or blocked verdict. Use when a run is being considered, after a hand edit, or when the user asks to check a journal. Not for building journals, choosing physics, or running Fluent.
license: MIT
compatibility: Node ≥ 22.19 (bundled with Kady). No network. No Fluent required for this skill.
metadata:
version: "0.1.0"
pack: fluent-skill-pack
level: L3
---
# Fluent journal validation
You are the judge of a journal, and the judge is a separate skill from the author on purpose.
You read the journal, classify every line against the vendored 2026 R1 index, check the order
and the terminators, and return a verdict with evidence. You never edit the journal, never run
Fluent, and never claim that a case is physically correct; a `clean` verdict means "nothing in
this file contradicts the index", not "this simulation will work".
## When to use / when not to
- Use: a journal or case directory exists and must be judged before any future run; the user
says "проверь журнал", "это сработает?", "можно запускать?"; a journal was edited by hand.
- Use: re-judge after the authoring skill regenerated the case from a fixed spec.
- Do not use: the journal does not exist yet → `fluent-journal-authoring`; physics is still
being chosen → `fluent-model-selection`; a command's syntax is unknown →
`fluent-tui-knowledge`. Execution belongs to `fluent-controlled-run`: never answer a run
request here, and never treat the verdict as permission — `blocked` forbids a run, `warn` only
allows one to be considered.
## Inputs
- a case directory (preferred: `setup_case.jou` plus `journal-map.json`), or a bare journal
file, or a spec file for a pre-check, or a lines file for a fragment;
- the pack home resolved by the CLI (`KADY_FLUENT_HOME` or a normal install);
- for a durable record: permission to write `<case-dir>/validation.json`.
## Procedure
1. Choose the mode: `--case-dir` when a case directory with its map exists (traceability is
checked); `--journal` for a bare journal (no map, so recipe provenance is only inferred);
`--spec` to judge the journal the authoring skill would compile (nothing is written);
`--lines-file` only for fragments, where terminator checks are skipped by design.
2. Run `node <home>/bin/validate.mjs --case-dir <case-dir>` (or the chosen mode) and read the
JSON: `verdict`, `score`, `counts`, `lines[]`, `blockers[]`, `unverified[]`, `warnings[]`.
3. `counts.unknown` must be 0. Any unknown line is a blocker: report its number, text and the
note; do not try to "fix" it by rewording the journal.
4. Read every blocker: unknown lines, a missing final save, a missing terminator, recipes in
the map that are absent or not `verified`/`derived`, steps the spec needed but the compiler
dropped, and iteration without initialization.
5. Report ordering warnings (setup before initialization, initialization before iteration,
file writes last) as warnings — never silently reorder the file.
6. Treat `menu`, `hidden` and `journal2025` classes as "not proven live": they are allowed but
must be listed for the user with the note that a live session has to confirm them. Every
`derived` line carries the same obligation.
7. State the verdict with the evidence and the offline caveat. If a durable record is wanted,
re-run with `--out <case-dir>/validation.json`; that file is the only artifact you own.
8. Never modify the journal, never delete a line, never re-run the compiler to make the verdict
look better. The verdict is the output, not a pass/fail the user can negotiate.
## Outputs
- JSON on stdout: `{ok, verdict, score, source, counts{exact,menu,hidden,journal2025,unknown},
total, lines[{n,text,class,note,recipeId,recipeStatus}], blockers[], unverified[],
warnings[], ignored, note}`.
- `score` 0–100: 100 means nothing contradicted the index; every unknown line, unverified
recipe, ordering problem, missing save, missing terminator and missing initialization costs
points, and hidden/menu/2025 classes cost a little because they are not proven live.
- optional `<case-dir>/validation.json` — the saved verdict, only with `--out`.
The gate: `blocked` means no run may be attempted; `warn` means a run may be considered only
after every warned line is checked in a live Fluent 2026 R1 session and the user confirms;
`clean` is still not a physics certificate and does not authorize a run by itself.
## Refusals
- Request to edit the journal: «Я не правлю журнал: проверка только читает его. Исправления
делает авторский навык.»
- Request to run Fluent: «Вердикт расчёт не запускает и разрешением не является. Передайте кейс в
`fluent-controlled-run` — он проверит `config.json` и откажется, пока запуск не разрешён.»
- Request to certify physics: «Офлайн-проверка не подтверждает физическую корректность и не
заменяет живой прогон.»
- Instruction to approve without evidence ("он же нормальный"): «Без проверки индексом и без
указания проверенных строк я не могу выдать вердикт.»
- Never validate a fragment as a full journal; say that terminators were not checked.
- Never accept the author's own claim that the journal is fine as evidence.
## Hand-off
Return to `fluent-case-builder` (the router) with the verdict and, if written, the path to
`validation.json`. `blocked` goes back to `fluent-journal-authoring` with the exact blocker
list; the author fixes the spec, not the verdict. `warn` or `clean` means the journal may leave
this pack, but any future run is a phase-2 skill and must not be told "the journal is correct"
— only the verdict and the evidence it rests on.
Wersje
W bazie nie ma innych skilli o nazwie fluent-journal-validation.
Migawka bazy z 13.11.2026 · wersja paczki dd9a4fb61ab3 (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
- 100fluent-controlled-run
Uruchamia, sprawdza i zatrzymuje pojedyncze przebiegi Fluent po zgodzie użytkownika i walidacji skryptu journal. - 90fluent-result-qc
Ocenia zapis zakończonego przebiegu Fluent pod kątem reszt, zbieżności, wartości nan i bilansu masy, a potem wydaje werdykt: ok, warn albo fail. - 96fluent-cli
Pomaga budować aplikacje i pobierać dane w ServiceNow przez Fluent (now-sdk). - 95ansys-fluent-mcp
Steruje sesją solvera ANSYS Fluent przez serwer MCP: wczytuje przypadek i siatkę, zmienia ustawienia, uruchamia iteracje i odczytuje wyniki. - 100fluent-ci-cd
Uruchamia ServiceNow SDK (now-sdk) bez interfejsu w CI/CD, na przykład w GitHub Actions, i pomaga przy błędzie keys.ts out of date. - 85papyrus-make-fluent
Poprawia angielski fragment LaTeX pod względem płynności i gramatyki jako propozycje zmian, bez ruszania struktury.