prompt-engineering
/prompt-engineering
Pomaga tworzyć i ulepszać agentów, skille, komendy slash oraz instrukcje CLAUDE.md: pisanie opisu uruchamiającego, struktura promptu systemowego, zakres narzędzi i stopniowe ujawnianie wiedzy.
- Autor
- alisinadevelo ↗
- Repozytorium
- md-files ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 03.10.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 2
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 03.10.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Pomaga tworzyć i ulepszać agentów, skille, komendy slash oraz instrukcje CLAUDE.md: pisanie opisu uruchamiającego, struktura promptu systemowego, zakres narzędzi i stopniowe ujawnianie wiedzy.
Oryginalny opis (EN)
Use when authoring or improving Claude Code agents, skills, slash commands, or CLAUDE.md instructions. Covers how to write a triggering description, structure a system prompt, scope tools, and apply progressive disclosure so the model actually uses what you build. See PATTERNS.md for operating-discipline techniques distilled from production agent prompts.
Jak działa
- Descriptions decide whether you get loaded
- System prompts: role, method, output, boundaries
- Scope tools to the job
- Progressive disclosure (skills)
- Calibrate, don't overclaim
- Operating-discipline patterns
- Test it
Podgląd SKILL.md
---
name: prompt-engineering
description: >-
Use when authoring or improving Claude Code agents, skills, slash commands, or
CLAUDE.md instructions. Covers how to write a triggering description, structure a
system prompt, scope tools, and apply progressive disclosure so the model
actually uses what you build. See PATTERNS.md for operating-discipline techniques
distilled from production agent prompts.
---
# Prompt Engineering for Claude Code
The artifacts in this toolkit — agents, skills, commands, instructions — are prompts.
They only help if the model (a) loads them at the right moment and (b) can act on them.
These two failure modes drive every guideline below.
## Descriptions decide whether you get loaded
For agents and skills, the `description` is the *only* thing the model sees when deciding
whether to invoke. Write it for triggering, not for humans:
- Start with **"Use when…"** and name the concrete situations, symptoms, and keywords a
user would actually say.
- Include 1–2 example triggers. Specificity beats elegance.
- State what it's for *and what it's not* if there's a near neighbor (e.g. "for
design, not implementation").
- A vague description ("helps with code") never fires. A precise one fires reliably.
## System prompts: role, method, output, boundaries
Structure the body of an agent/skill as:
1. **Role** — who it is and what it optimizes for, in one or two sentences.
2. **Method** — the concrete steps or checklist it follows. Numbered, specific.
3. **Output format** — the exact shape you want back. Models follow a template well.
4. **Boundaries** — what *not* to do, and what to do when uncertain (ask vs. assume).
Write positive instructions ("do X") over long lists of prohibitions. Show the format
with a small example rather than describing it abstractly.
## Scope tools to the job
Give an agent only the tools it needs. A reviewer or auditor is read-only (Read, Grep,
Glob, Bash) — no Edit. A refactorer needs Edit. Narrow tools reduce mistakes and make the
agent's contract obvious.
## Progressive disclosure (skills)
Keep `SKILL.md` lean — the high-frequency guidance. Push exhaustive references
(checklists, catalogs, schemas) into sibling files the model loads only when needed, and
point to them by name. This keeps the always-loaded surface small while making depth
available on demand. (See `code-review-rubric` and `refactoring-catalog` here for the
pattern.)
## Calibrate, don't overclaim
Tell the model to express uncertainty, prefer evidence over assertion, and stop and ask
when a decision is genuinely the user's. An agent that fabricates confidence is worse
than one that says "I'm not sure — here's what would confirm it."
## Operating-discipline patterns
The sections above cover *structure* — how to make an agent load and be actionable. For the
*behavioral* techniques that separate a sharp agent from a vague one (parallel tool use,
read-before-edit, verify-with-evidence, autonomy and stop conditions, output contracts,
delegation briefing), load **PATTERNS.md**. It distills what production coding agents
converge on, so you can state the right discipline in each agent's Method/Boundaries.
## Test it
After writing, sanity-check the trigger: would the description fire on a realistic
request? Would it *over*-fire on unrelated ones? Tune the description until it's sharp.
Pliki w paczce (2)
- PATTERNS.md 5 KB
- SKILL.md 3 KB
Wersje
| Repozytorium | Wynik | Licencja |
|---|---|---|
| Ta strona · alisinadevelo/md-files | 100 | MIT |
| Muvon/octomind-tap | 100 | Apache-2.0 |
| TerminalSkills/skills | 100 | Apache-2.0 |
| ericrisco/rsc-harness | 100 | MIT |
| wrongstack/wrongstack | 100 | MIT |
| rootcastleco/rei-skills | 100 | MIT |
| FrancoStino/opencode-skills-collection | 100 | MIT |
| davila7/claude-code-templates | 100 | MIT |
| it-bens/ai-tools | 100 | MIT |
Migawka bazy z 13.11.2026 · wersja paczki 2e1c8672ad78 (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
- 90Prompt Brevity Review
Recenzuje prompty i instrukcje AI pod kątem zwięzłości i jasności, w tym skille, CLAUDE.md i komendy slash. - 96writing-skills
Pomaga tworzyć i ulepszać skille agentów AI, czyli instrukcje dla agentów. - 100skill-authoring
Pomaga napisać nowy skill dla agenta AI: zakres, opis dla niezawodnego wyzwalania, progresywne ujawnianie i różnica między skillem a promptem. - 100create-agent-skills
Doradza przy tworzeniu skilli i komend slash dla Claude Code: pliki SKILL.md, struktura skilla i dobre praktyki. - 100claude-code-slash-commands
Pomaga tworzyć i konfigurować własne komendy slash w Claude Code: argumenty, wykonywanie bash, struktura i dobre praktyki. - 92create-agent-skills
Doradza przy tworzeniu i doskonaleniu skilli Claude Code: pliki SKILL.md, struktura, stopniowe ujawnianie, przepływy pracy, wzorce walidacji i formatowanie XML.