agentget

agent
SUMMARY

Install AI agents, instructions, skills, and rules from GitHub repos into your project/globally

README.md

agentget

npm version
npm downloads
CI
License: ISC

Install AI agents, instructions, skills, and rules from GitHub repos into your project.

Installation

# npm
npm install -g agentget

# bun
bun add -g agentget

Or use without installing via npx / bunx:

npx agentget add owner/repo
bunx agentget add owner/repo

Usage

# Both of these work:
npx agentget add owner/repo
npx agent-get add owner/repo

# Install all agents (default)
npx agentget add owner/repo

# Install everything (agents, skills, instructions, rules)
npx agentget add owner/repo --all

# Install specific agent + all skills/instructions/rules
npx agentget add owner/repo --agent code-reviewer

# Install only skills
npx agentget add owner/repo --skills-only

# Install only instructions
npx agentget add owner/repo --instructions-only

# Install only rules
npx agentget add owner/repo --rules-only

# Install specific agent only (no skills/instructions/rules)
npx agentget add owner/repo --agent code-reviewer --agents-only

# Use alternative directory names (for repos with non-standard structure)
npx agentget add VoltAgent/awesome-claude-code-subagents --agents-dir categories

List Installed

# List installed agents (default)
npx agentget list

# List everything (agents, skills, instructions, rules)
npx agentget list --all

# List only agents
npx agentget list --agents-only

# List only skills
npx agentget list --skills-only

# List only instructions
npx agentget list --instructions-only

# List only rules
npx agentget list --rules-only
Flag Behavior
(none) Lists agents only (default)
--all Lists everything
--agents-only Lists agents only (explicit)
--skills-only Lists skills only
--instructions-only Lists instructions only
--rules-only Lists rules only

Filtering

Flag Behavior
(none) Installs agents only (default)
--all Installs everything
--agent <name> Installs specified agent + all skills/instructions/rules
--agents-only Installs agents only (explicit)
--skills-only Installs skills only
--instructions-only Installs instructions only
--rules-only Installs rules only

Alternative Directory Names

For repos with non-standard structure, specify custom directory names:

Flag Default Usage
--agents-dir <name> agents --agents-dir categories
--skills-dir <name> skills --skills-dir my-skills
--instructions-dir <name> instructions --instructions-dir docs
--rules-dir <name> rules --rules-dir policies

Note: Agents directory scanning is recursive — finds .md and .agent.md files in all subdirectories.

Frontmatter Handling

YAML frontmatter (content between --- delimiters at the start of files) is automatically stripped during installation. This ensures compatibility with tools that don't support frontmatter or have different frontmatter requirements.

Example:

---
name: ui-designer
tools: Read, Write, Edit
---

You are a UI designer...

Becomes:

You are a UI designer...

What it installs

Content is discovered from these patterns in the target repo:

Type Pattern
Agents agents/*.agent.md
Instructions instructions/*.instructions.md
Skills skills/*/SKILL.md (whole folder)
Rules rules/*.rules.md
Plugins plugins/*/ (expanded recursively)

Where files go

agentget supports 41 AI coding tools. Content is always written to a canonical location first, then symlinked to detected tool directories.

Supported vs Detected

  • Supported — all 41 tools agentget knows how to install into. Run agentget targets to list them.
  • Detected — the subset whose config directories exist on your machine right now. Only detected targets receive symlinks, avoiding unused dot-directories.
  • Canonical — 10 tools that read .agents/ directly and are always active. No symlink needed.

Canonical (source of truth)

.agents/agents/<name>.agent.md
.agents/instructions/<name>.instructions.md
.agents/skills/<name>/
.agents/rules/<name>.rules.md

Canonical readers (always active)

These tools read .agents/ directly — no symlinks needed:

AMP, Cline, Codex, Cursor, Gemini CLI, GitHub Copilot, Kimi Code CLI, OpenCode, Replit, Universal

Project targets

Symlinks created when the tool's project directory is detected:

Tool Path
AdaL .adal/{agents,instructions,skills,rules}/
Antigravity .agent/{agents,instructions,skills,rules}/
Augment .augment/{agents,instructions,skills,rules}/
Claude Code .claude/{agents,instructions,skills,rules}/
CodeBuddy .codebuddy/{agents,instructions,skills,rules}/
Command Code .commandcode/{agents,instructions,skills,rules}/
Continue .continue/{agents,instructions,skills,rules}/
Cortex Code .cortex/{agents,instructions,skills,rules}/
Crush .crush/{agents,instructions,skills,rules}/
Droid .factory/{agents,instructions,skills,rules}/
Goose .goose/{agents,instructions,skills,rules}/
iFlow CLI .iflow/{agents,instructions,skills,rules}/
Junie .junie/{agents,instructions,skills,rules}/
Kilo Code .kilocode/{agents,instructions,skills,rules}/
Kiro CLI .kiro/{agents,instructions,skills,rules}/
Kode .kode/{agents,instructions,skills,rules}/
MCPJam .mcpjam/{agents,instructions,skills,rules}/
Mistral Vibe .vibe/{agents,instructions,skills,rules}/
Mux .mux/{agents,instructions,skills,rules}/
Neovate .neovate/{agents,instructions,skills,rules}/
OpenClaw {agents,instructions,skills,rules}/ (requires .openclaw/.clawdbot/.moltbot marker)
OpenHands .openhands/{agents,instructions,skills,rules}/
Pi .pi/{agents,instructions,skills,rules}/
Pochi .pochi/{agents,instructions,skills,rules}/
Qoder .qoder/{agents,instructions,skills,rules}/
Qwen Code .qwen/{agents,instructions,skills,rules}/
Roo Code .roo/{agents,instructions,skills,rules}/
Trae .trae/{agents,instructions,skills,rules}/
Trae CN .trae/{agents,instructions,skills,rules}/
Windsurf .windsurf/{agents,instructions,skills,rules}/
Zencoder .zencoder/{agents,instructions,skills,rules}/

Global targets

Global symlinks are only created when the tool's config directory already exists (or an env var like CLAUDE_CONFIG_DIR / CODEX_HOME points to it). This avoids spraying unused dot-directories across your home directory.

Tool Global path
AdaL ~/.adal/
AMP / Kimi Code CLI / Replit / Universal ~/.config/agents/
Antigravity ~/.gemini/antigravity/
Augment ~/.augment/
Claude Code ${CLAUDE_CONFIG_DIR:-~/.claude}/
Cline ~/.agents/
CodeBuddy ~/.codebuddy/
Codex ${CODEX_HOME:-~/.codex}/
Command Code ~/.commandcode/
Continue ~/.continue/
Cortex Code ~/.snowflake/cortex/
Crush ~/.config/crush/
Cursor ~/.cursor/
Droid ~/.factory/
Gemini CLI ~/.gemini/
GitHub Copilot ~/.copilot/
Goose ~/.config/goose/
iFlow CLI ~/.iflow/
Junie ~/.junie/
Kilo Code ~/.kilocode/
Kiro CLI ~/.kiro/
Kode ~/.kode/
MCPJam ~/.mcpjam/
Mistral Vibe ~/.vibe/
Mux ~/.mux/
Neovate ~/.neovate/
OpenClaw / ClawdBot / MoltBot ~/.openclaw/ (auto-detects .clawdbot or .moltbot)
OpenCode ~/.config/opencode/
OpenHands ~/.openhands/
Pi ~/.pi/agent/
Pochi ~/.pochi/
Qoder ~/.qoder/
Qwen Code ~/.qwen/
Roo Code ~/.roo/
Trae ~/.trae/
Trae CN ~/.trae-cn/
Windsurf ~/.codeium/windsurf/
Zencoder ~/.zencoder/

All global paths contain {agents,instructions,skills,rules}/ subdirectories.

Builtin Agents

agentget ships a curated set of agents you can install directly from this repo:

npx agentget add joeyism/agentget --agent <name>
Agent Install command Description
Sisyphus-Junior --agent sisyphus-junior Focused task executor that completes delegated atomic tasks without spawning other agents. It follows existing codebase patterns, verifies its own work with diagnostics and build checks, and never refactors while fixing.
Atlas --agent atlas Master orchestrator that drives a full todo list to completion by dispatching tasks to specialized agents in parallel waves. It verifies each task's evidence before proceeding and runs a four-agent Final Verification Wave at the end.
Prometheus --agent prometheus Strategic planning consultant that interviews you to clarify requirements, then generates a structured work plan in .sisyphus/plans/. It never writes code — it only plans, and hands off to executors via /start-work.
Oracle --agent oracle Read-only high-IQ reasoning specialist for architecture decisions, hard debugging, and post-implementation review. Consult it after 2+ failed fix attempts or before committing to a complex design.
Metis --agent metis Pre-planning consultant that classifies intent, surfaces hidden requirements, and produces guardrail directives for Prometheus before planning begins. Prevents AI over-engineering and scope creep.
Momus --agent momus Work plan reviewer that checks plans for executability and valid references. Approves by default — it only blocks on true blockers (missing files, zero-context tasks, contradictions), never on stylistic concerns.
Explore --agent explore Contextual grep specialist for your own codebase. Fire multiple instances in parallel to find where patterns are implemented, how modules are structured, and which files are relevant — returns absolute paths with structured results.
Librarian --agent librarian External reference specialist that finds official documentation, open-source implementations, and best practices for any library. Every claim is backed by a GitHub permalink to the exact source line.
Hephaestus --agent hephaestus Autonomous deep worker that explores the codebase and external docs thoroughly before writing a single line, then completes complex multi-file implementations end-to-end without check-ins.
Multimodal Looker --agent multimodal-looker Media file interpreter that extracts specific information from PDFs, images, and diagrams. Use it when you need analyzed data from a file rather than its raw contents — it saves context tokens for the main agent.

Yorumlar (0)

Sonuc bulunamadi