convez.ai
Skill
tripleyak

prompting-skill

tripleyak/prompting-skill

Ver no GitHub
InicialPluginScore · de 0 a 100

Agent skill for Claude Code & Codex: write, review, and upgrade prompts for frontier coding models (Claude Fable 5 patterns, paste-ready blocks, pattern library)

Instalação

$npx convez-ai tripleyak/prompting-skill

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
28%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

README

A tradução para o português deste README ainda está na fila.

prompting — an agent skill for writing better prompts

A skill for Claude Code and Codex CLI that helps the agent write, review, and upgrade prompts for frontier coding models — especially Claude Fable 5, where old prescriptive prompting habits actively degrade output.

What's in it

FileContents
SKILL.mdEntry point: workflow, core prompting principles (intent-first, define-done, grounded progress claims, bounded autonomy)
fable-5.mdFable 5 behavior deltas vs Opus-class models + 12 paste-ready prompt blocks keyed by symptom (overplanning, scope creep, fabricated progress, early stopping, context anxiety…) + anti-patterns
patterns.mdComposable pattern library: overnight delegation, architecture planning, visual verification loops, root-cause analysis, evidence-based research, spec-driven builds, plus a 5-part prompt skeleton

Install

Claude Code:

git clone https://github.com/tripleyak/prompting-skill ~/.claude/skills/prompting

Codex CLI:

git clone https://github.com/tripleyak/prompting-skill ~/.codex/skills/prompting

Restart your session afterwards — skills load at session start.

Then ask things like "improve this prompt", "write me a prompt for an overnight run", or "why does the model keep refactoring things I didn't ask for?"

Notes

  • SKILL.md references two optional local files (~/.claude/rules/model-profiles.md, ~/.claude/rules/loops.md) from the author's setup. The skill works fine without them — they cover compensations for weaker executor models and loop/verifier design. Delete those lines or write your own.
  • Sources: Anthropic's Prompting Claude Fable 5 guide, the skill authoring best practices, and Every.to's Claude Fable 5 prompt library.

License

MIT

Ver o original no GitHub

Bom para

Relacionados