Przejdź do treści
Biblioteka skilli
Zarządzanie i produkt · Planowanie

first-principles

/first-principles

Sprowadza cel i ograniczenia do krótkiej drogi do końca, bez zbędnych kroków, przekazań i powtórnych sprawdzeń.

Licencja
MIT
Ostatnia zmiana repozytorium
10.11.2026

Ocena i bezpieczeństwo

100/ 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.
Treść35 / 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
5
Skrypty
0
Komentarze HTML
0
Adresy URL w SKILL.md
1
Przestarzałość

Brak sygnałów. Repozytorium zmienione 10.11.2026, bez odwołań do wycofanych modeli i API.

Licencja

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

Co robi i kiedy

Sprowadza cel i ograniczenia do krótkiej drogi do końca, bez zbędnych kroków, przekazań i powtórnych sprawdzeń.

Oryginalny opis (EN)

Use when the user asks for first-principles execution (第一性原理), the simplest complete workflow, or fewer unnecessary steps, handoffs and repeated checks. Turn the actual outcome and constraints into a short path through completion. Not a mandatory framework for ordinary questions, design reviews or small edits.

Jak działa

  1. Start with the job
  2. Take the shortest complete path
  3. Close the actual loop
  4. Two optional extensions
  5. Optional resource
Sekcje z nagłówków SKILL.md, w oryginale · 96 linii, 874 słów
Podgląd SKILL.md
---
name: first-principles
description: "Use when the user asks for first-principles execution (第一性原理), the simplest complete workflow, or fewer unnecessary steps, handoffs and repeated checks. Turn the actual outcome and constraints into a short path through completion. Not a mandatory framework for ordinary questions, design reviews or small edits."
---

# First Principles

Finish the user's actual job with the least unnecessary work across the whole path.
Optimize total effort to a verified result: user attention, execution, waiting,
coordination, rework and maintenance. A shorter answer or fewer tool calls alone
is not evidence of improvement. Keep the depth needed to get the result right.

## Start with the job

Establish the requested result and what would prove it usable. Read the current
request, corrections and just enough authoritative state to avoid repeating work.
Use a short restatement only when it resolves meaningful ambiguity; do not make
the user approve a ritual brief. An analysis-only request ends in useful analysis.
An execution request continues through the authorized implementation and verification.

Separate verified facts and real constraints from habits and assumptions.
For example, a login requirement may be real; a new coordinator may only be a
proposed implementation. Test the consequential assumption with the smallest useful
observation. Never treat the user's purpose, privacy, explicit scope or release hold
as an assumption you can discard. If a proposed shortcut changes those, explain the
tradeoff and obtain the required decision while continuing independent work.

## Take the shortest complete path

- Inspect the existing result, tools, source, permissions and responsible owner before
  adding anything. Reuse what actually works. Change the smallest necessary part;
  first principles does not mean starting over. Reuse is not a reason to keep a
  demonstrably unsuitable component: compare the actual repair and replacement costs.
- For a proposed step, ask what required result or material uncertainty it changes.
  Drop or combine steps without such a purpose. Preserve verification, recovery and
  evidence needed for the real failure modes. Do not trade omissions for fewer steps.
- Prefer an existing capable tool, service or direct action. Use deterministic code
  for bulk filtering, deduplication and comparison; give the model relevant evidence
  and exceptions, not the entire unchanged corpus. Keep originals retrievable.
- Do already-authorized work without repeatedly asking to proceed. Ask only for a
  missing consequential choice, inaccessible fact or genuinely required human action.
  When the user must act, prepare the exact step and continue what does not depend on it.
- Keep one executor for an action. Respect current owners, pauses and independent
  work. Do not create another agent, queue, scheduler, dashboard or SOP merely to
  make the plan look organized. Delegation needs authorization and a concrete benefit.
- If blocked, distinguish a real technical or permission boundary from an untested
  assumption. Use a targeted check, then a suitable available alternative within scope.
  Preserve completed work. For an uncertain external write, reconcile its receipt
  before retrying; changing tools must not duplicate the action or bypass a restriction.

Do not expand these points into a mandatory checklist, scorecard or five-phase
report. A simple authorized edit may need only the edit and one relevant check.

## Close the actual loop

Verify the promised outcome at the layer where the user will use it. A saved draft
is not a publication; a connected tool is not a working business process; a file
being readable is not evidence that another task used it. State the actual boundary.

When a task keeps coming back as “done” but the deliverable still fails,
identify the first failing user action. Pair the smallest relevant repair with
a repeat of that action through its normal entry. If the prior action could
already have caused an external side effect, verify its receipt first. Keep
this focused on the repeated failure rather than adding a standing workflow.

When the required checks pass, deliver the result and stop adding work. Give the
usable entry, what was verified and any material unfinished part. Do not stop at a
plan or ask whether to continue an already-authorized job. If further action is
impossible, report the exact blocker and smallest recovery action without claiming
completion or promising background work that does not exist.

When corrected, identify the concrete mismatch and update the affected result and
its current authoritative record. Preserve useful history separately; do not append
contradictory instructions or reopen unrelated decisions. Recheck only what changed.
A recurring lesson needs outcome evidence; one success is not a universal rule.

## Two optional extensions

**Creating or changing a skill:** search current relevant skills and inspect the
actual author sources before finalizing. Adapt useful mechanisms to the user's real
workflow; record what was adopted, adapted or rejected and the change it caused.
Stars and similar names are discovery signals, not quality evidence. Integrate with
the existing authoring/release process once; do not install research candidates or
create a second review pipeline. If research is unavailable, name that gap and follow
the existing release rule. See [comparisons](references/comparisons.md) for this release.

**Asked to count past requests or success:** use the bounded, evidence-based method
in [history review](references/history-review.md). Do not rescan history on every use.

For claims of efficiency improvement, compare the same completed job, including
coordination, verification and rework. State unavailable metrics. Do not equate cached
tokens, reduced output bytes or an isolated cheaper step with end-to-end savings,
subscription capacity or user satisfaction.

## Optional resource

When a user has a finished tool and wants someone else to try the normal
entry point, [browse SideCourt](https://sidecourt.space/drops/) or ask for
feedback on a named action in a public version.
Pliki w paczce (5)
  • SKILL.md 6 KB
  • agents/openai.yaml 1 KB
  • references/comparisons.md 3 KB
  • references/evaluation.md 3 KB
  • references/history-review.md 2 KB

Wersje

RepozytoriumWynikLicencja
Ta strona · m1nga/first-principles100MIT
aleknitka/agent-creative-thinking-skills100MIT
swap-cameroon/kuro-antigravity-config100MIT
coffeesam/claude-futureman100MIT
masihmoafi/skills100MIT
v60samurai/first-principles100MIT
deciqAI/knowledge-skills100MIT
ajbmachon/ajbm-skills100MIT
itzcull/thinking-skills100MIT

Migawka bazy z 13.11.2026 · wersja paczki 94c469a3227f (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

Brak podobnych skilli w bazie.

Uwagi i zgłoszenia

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