convez.ai
Agente
Hbrinj

claude-prompt

Hbrinj/claude-prompt

Ver no GitHub
InicialPluginScore · de 0 a 100

Centralised collection of reusable Claude Code agents, shared across multiple repositories

Instalação

$npx convez-ai Hbrinj/claude-prompt

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 4 meses
Versão
—
Issues abertas
0
Contribuidores
3

README

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

Claude Agents & Skills

A centralised collection of reusable Claude Code agents and skills, designed to be shared across multiple repositories.

Overview

This repository contains custom Claude Code subagents and slash-command skills that can be installed globally or per-project. Each agent and skill is a self-contained Markdown file.

Structure

agents/
├── README.md         # Agent catalogue and usage notes
└── *.md              # Individual agent definitions
skills/
├── README.md         # Skills catalogue and usage notes
└── *.md              # Individual skill definitions
SYSTEM_PROMPT.md      # Coordinator workflow skeleton — source of truth with skills/implement-feature.md (committed)
CLAUDE.md             # Repo conventions for Claude when working in this repo

Workflow source of truth

The canonical, version-controlled coordinator workflow is split across two files: SYSTEM_PROMPT.md (always-loaded skeleton — role, non-negotiable rules, Step 1 → Understand/Specify/Slice and Step 2 → Implement outline) and skills/implement-feature.md (Step 2 procedure, agent/skill routing index, checkpoint format, feature-log schema — loaded on demand). Two derived copies of SYSTEM_PROMPT.md exist downstream:

  • Consumer repos' CLAUDE.md — kept in sync automatically by the sync-upstream skill (skills/sync-upstream.md), which writes the latest SYSTEM_PROMPT.md content into a guarded block bounded by <!-- SYSTEM_PROMPT:START --> / <!-- SYSTEM_PROMPT:END --> markers. Anything in the consumer repo's CLAUDE.md outside those markers is preserved.
  • Your ~/.claude/CLAUDE.md — Claude Code's global instructions file. Kept in sync manually — sync-upstream does NOT touch it. After editing SYSTEM_PROMPT.md, copy the new content into ~/.claude/CLAUDE.md yourself if you want the active session and other "raw" sessions (no sync-upstream installed) to pick up the change immediately.

When changing the coordinator workflow, ALWAYS edit SYSTEM_PROMPT.md and/or skills/implement-feature.md in this repo — NEVER edit a derived copy (~/.claude/CLAUDE.md or any consumer repo's CLAUDE.md) directly.

Editing a derived copy alone is upside-down: the change is not version-controlled, is not shared with other machines or users, and is silently overwritten the next time sync-upstream runs in that consumer repo. The fix is one-way: edit SYSTEM_PROMPT.md here, commit, then propagate (sync-upstream for consumer repos, manual copy for ~/.claude/CLAUDE.md).

If you're unsure where a coordinator-shaped change belongs, default to the source-of-truth pair: SYSTEM_PROMPT.md for skeleton/rules changes, skills/implement-feature.md for Step 2 procedure changes.

The same source-of-truth rule applies to agents/*.md and skills/*.md — the repo is the source; ~/.claude/agents and ~/.claude/commands are derived (typically symlinked by sync-upstream, see Usage).

Usage

Global install (available in every repo)

# Symlink or copy individual agents
cp agents/<agent-name>.md ~/.claude/agents/

# Or symlink the whole directory (changes here apply everywhere)
ln -s "$(pwd)/agents" ~/.claude/agents

Per-project install

mkdir -p .claude/agents
cp path/to/claude-prompt/agents/<agent-name>.md .claude/agents/

Invoke an agent

Once installed, reference any agent in Claude Code:

Use the <agent-name> agent to …

Or from the CLI:

claude --agent <agent-name> "your task here"

Skills

Skills are reusable slash commands (/skill-name) that appear in Claude Code's command palette.

Global install

cp skills/<skill-name>.md ~/.claude/commands/

# Or symlink the whole directory
ln -s "$(pwd)/skills" ~/.claude/commands

Per-project install

mkdir -p .claude/commands
cp path/to/claude-prompt/skills/<skill-name>.md .claude/commands/

Invoke a skill

/skill-name

Adding a new skill

  1. Create a new .md file in skills/.
  2. Write the prompt body (no frontmatter required, but description: is recommended).
  3. Open a PR — the skill becomes available to all repos on merge.
  4. Update skills/README.md with a one-line entry.

Adding a new agent

  1. Create a new .md file in agents/.
  2. Add the required frontmatter (name, description, tools).
  3. Write the system prompt body.
  4. Open a PR — the agent becomes available to all repos on merge.

Agent template

---
name: agent-name
description: One-line description of when Claude should use this agent.
tools: Read, Edit, Bash   # comma-separated list of tools this agent may use
---

You are a specialist in …

## Responsibilities
- …

## Constraints
- …

Guard hooks

scripts/hooks/ contains PreToolUse hooks that mechanically enforce two workflow non-negotiables: guard-main-edit.sh denies Edit/Write/NotebookEdit while the target's repo is on main/master, and guard-push-main.sh denies git push to main/master. install.sh --global symlinks ~/.claude/hooks → <clone>/scripts/hooks and prints the settings.json block to add manually (it never edits settings.json). Both hooks require jq (preinstalled on recent macOS; otherwise brew install jq) and fail open when jq is missing or stdin is malformed. Per-path opt-outs for the edit guard live in ~/.claude/hooks-exceptions. Full contract: scripts/hooks/README.md.

Contributing

  • Keep each agent and skill focused on a single responsibility.
  • For agents: prefer narrow tool lists — only grant the tools an agent genuinely needs.
  • Update agents/README.md or skills/README.md with a one-line entry for every new addition.
Ver o original no GitHub

Bom para

Relacionados