Przejdź do treści
Biblioteka skilli
Programowanie · Analiza

decide

/decide

Przygotowuje rozstrzygnięcie ważnej, złożonej albo trudnej do cofnięcia decyzji technicznej, zanim trafi ona do osoby odpowiedzialnej.

Repozytorium
skills ↗
Licencja
MIT
Ostatnia zmiana repozytorium
03.11.2026

Ocena i bezpieczeństwo

90/ 100Klasa: dobry
Format25 / 25
Frontmatter z nazwą i opisem, poprawna nazwa, opis do 1024 znaków, nazwa zgodna z folderem.
Odkrywalność25 / 25
Opis mówi, kiedy użyć skilla, ma sensowną długość i jest w 3. osobie.
Treść25 / 35
Co najmniej 150 słów i do 500 linii, nagłówki, listy kroków, przykłady albo kod.
Higiena15 / 15
Bez zaślepek typu TODO, bez ścieżek Windows, z licencją.
Ryzykobrak sygnałów w skanie
Pliki w paczce
2
Skrypty
0
Komentarze HTML
0
Adresy URL w SKILL.md
0
Przestarzałość

Brak sygnałów. Repozytorium zmienione 03.11.2026, bez odwołań do wycofanych modeli i API.

Licencja

MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.

Co robi i kiedy

Przygotowuje rozstrzygnięcie ważnej, złożonej albo trudnej do cofnięcia decyzji technicznej, zanim trafi ona do osoby odpowiedzialnej.

Oryginalny opis (EN)

Use when facing an important, complex, or hard-to-reverse technical decision, or whenever a technical decision would otherwise be put to the user.

Jak działa

  1. Objective
  2. Workflow
  3. Human versus technical decisions
  4. Delegated decisions
  5. Report
  6. Boundaries
Sekcje z nagłówków SKILL.md, w oryginale · 101 linii, 775 słów
Podgląd SKILL.md
---
name: decide
description: Use when facing an important, complex, or hard-to-reverse technical decision, or whenever a technical decision would otherwise be put to the user.
---

# Decide

## Objective

Reach the best overall solution to an important or complex problem through
deliberate research and judgment, instead of jumping to the first plausible
fix. This skill combines `/research`, `/research-web`, `/options`,
`/criteria`, `/root-cause`, and `/approach`.
The deliverable is a well-grounded proposal, not an immediate implementation.

## Workflow

1. **Think hard.** Restate the problem or problems at hand, the goal, the
   constraints, and what a good outcome looks like. Name the open questions and
   assumptions before looking for answers. Do not settle on a solution yet.
2. **Research locally with `/research`.** Inspect the relevant code, configs,
   specs, history, and existing mechanisms until you understand how the system
   actually works and where the problem lives.
3. **Research externally with `/research-web`.** Look up current docs, upstream
   projects, prior art, known issues, and how others solved the same problem.
   Prefer primary sources. Invoking this skill directly counts as an explicit
   request for web research; a delegated decision scales it as described under
   "Delegated decisions".
4. **Generate candidates.** Draw several distinct approaches from the research,
   including the obvious one, a more conservative one, and a more thorough one
   when they differ meaningfully. Drop candidates the evidence already rules
   out.
5. **Compare with `/options`.** Rank the strongest candidates, at most
   three, with their pros, cons, and what each entails.
6. **Rate with `/criteria`.** Rate each remaining candidate on
   usefulness, soundness, consistency, and proportionality, and let any `Weak`
   or `Violated` rating reshape or eliminate it.
7. **Screen for hacks with `/root-cause`.** Check each
   remaining candidate for workarounds, symptom patches, and fixes layered on
   top of broken behavior instead of changing the code that owns it. Replace
   such candidates with the root-cause version, or drop them.
8. **Confirm with `/approach`.** Judge the leading candidate as the
   long-term architecture: root cause addressed, idiomatic for the codebase,
   scoped proportionately. If it falls short, revise it or promote the next
   candidate.
9. **Propose.** Recommend the best overall solution for each problem. When the
   problems are related, prefer one coherent solution over separate patches.

Iterate when a later step reveals something new: if judging exposes a gap in
the research or a better candidate, go back and close it before proposing.

## Human versus technical decisions

This is the definition other skills use when they decide who settles a
question.

- **Technical decisions** have a defensible answer an engineer can reach from
  the requirements, the codebase, and research: which mechanism, library,
  data model, module boundary, protocol, file layout, or algorithm to use, and
  how to meet a requirement the user already set.
- **Human decisions** rest on the user's intent, audience, money, risk
  appetite, taste, or priorities, with no technical winner, or have broad
  implications: reshaping architecture across many modules or hosts, changing
  the security or privacy model, touching the repository's migration
  boundaries, removing or replacing features, deleting data, costing money,
  or affecting anything outward-facing.

When a question sits between the two, treat it as human. A stated user
preference turns a technical choice into a given; do not reopen it. When a
technical choice turns on a human answer not yet given, that answer comes
first.

## Delegated decisions

By default this skill proposes. When the user or a calling skill delegates a
decision, such as `/gaps-settle`, `/unblock`, `/wizard`, an unattended `/goal`, or
an agent settling a technical question mid-task under the user's standing
instructions, settle each **technical** decision by adopting the recommendation,
record the decision and its one-line rationale where the caller keeps its
work (the plan, task, or report), and let the caller carry on. Still surface
**human** decisions instead of settling them; the caller decides how to ask
or record them. Scale the research to the decision: local research is enough
unless it is high priority or turns on external facts such as an upstream
API, a release, or a known issue, and a delegated decision is not by itself a
request for web research.

## Report

- The problem or problems as understood, in a sentence or two each.
- Key findings from local and web research, with file references and source
  links.
- The `/options` comparison and the `/criteria` ratings.
- The `/root-cause` findings, briefly, and the `/approach` verdict for the recommendation.
- The recommended solution or solutions, why they win, and the main risks or
  open decisions that remain for the user.

## Boundaries

- Propose; do not implement unless the user asked for implementation as well.
  A delegated decision is settled, not implemented; the caller implements.
- Keep research read-only: no edits, installs, or state-changing commands while
  investigating.
- Surface genuine open decisions, and every human decision, to the user
  rather than resolving them by guesswork.
Pliki w paczce (2)
  • SKILL.md 5 KB
  • agents/openai.yaml 1 KB

Wersje

RepozytoriumWynikLicencja
Ta strona · yuri-xyz/skills90MIT
vegardjorg/ravenseat100MIT
generous-corp/pulp100MIT
wrg32786/aigent-os100MIT
alirezarezvani/claude-skills100MIT
AO-Cyber-Systems/devflow-claude100MIT
DvJ4Y/shipwright100MIT
blockrunai/blockrun-mcp100MIT
Abilityai/cornelius100MIT

Migawka bazy z 13.11.2026 · wersja paczki 8164654dcaa6 (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

  • 100adr-writer
    Pisze lub zastępuje rekord decyzji architektonicznej, zwykle przy decyzji trudnej do odwrócenia.
  • 92requesting-code-review
    Sprawdza, czy po zakończeniu zadania lub ważnej funkcji praca spełnia wymagania, zanim trafi do scalenia.
  • 100agdr-decide
    Wymusza proces Agent Decision Record: analizuje opcje przy nietrywialnej decyzji technicznej, dokumentuje wybór w docs/agdr/ i dopiero potem wdraża rozwiązanie.
  • 100architecture-decision
    Zapisuje rekord decyzji architektonicznej z kontekstem, opcjami, decyzją i konsekwencjami, gdy wybór jest trudny do cofnięcia.
  • 100create-rfc
    Tworzy dokument Request for Comments dla istotnej zmiany technicznej, przeglądu projektu lub zebrania opinii zespołów o decyzji technicznej.
  • 100repo-memory
    Zapisuje i wyszukuje decyzje architektoniczne repozytorium jako ADR w plans/architecture, także przy zmianach trudnych do cofnięcia.

Uwagi i zgłoszenia

Wiadomość trafi do Bartosza Sokolińskiego. Adres e-mail jest potrzebny tylko do odpowiedzi. Szczegóły w polityce prywatności.