Skip to main content
Execute qualquer Skill no Manus
com um clique

writing-agent-prompts

Estrelas6
Forks3
Atualizado29 de junho de 2026 às 09:10

Use when writing or editing ANY PaperHub agent/LLM prompt YAML — BOTH a DEPLOYMENT prompt (backend/src/paperhub/llm/prompts/*.yaml system/user template) AND an EVAL/EXPERIMENT variant (backend/benchmark/agent/prompts/<stage>/<version>.yaml driven by the agent benchmark sweep) — a new agent/router/synthesizer/subagent stage, or any change to how a model is instructed. Fires whenever you are tempted to hand-write one prompt and ship it, tweak a single prompt line "obviously," or adopt a rewrite without comparing it against the current one across multiple queries. Experiment/benchmark variants are NOT exempt: they get the SAME discipline (>=2 variants x query set x judged comparison; concise, direct, minimal-token rewrites), never a one-off draft.

Instalação

Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.

SKILL.md
readonly