Metadata

name

manage-agents

description

Use Agentloom to find, create, import, update, sync, and delete agents, commands, skills, and MCP servers.

Manage Agentloom Resources

Use this skill for end-to-end Agentloom management across agents, commands, skills, and MCP servers.

When to Use

Use this when the user asks to:

• find an existing agent/command/skill/MCP server
• import resources from a source repo
• create new local resources from scratch
• update tracked imports
• sync resources to provider outputs
• delete imported sources or entities
• add/remove manual MCP server entries

Canonical Layout

.agents/
  agents/
    *.md
  commands/
    *.md
  skills/
    <skill>/
      SKILL.md
      references/
      assets/
  mcp.json
  agents.lock.json
  settings.local.json

Core Commands

• Aggregate: agentloom add|find|update|sync|delete
• Entity scoped: agentloom agent|command|skill|mcp <add|list|delete|find|update|sync>
• Manual MCP server mode: agentloom mcp server add|list|delete

Common selectors/flags:

• --agents, --commands, --skills, --mcps
• --selection-mode <all|sync-all|custom>
• --subdir <path>
• --local / --global
• --yes

Workflow

1) Capture intent

Extract:

1. target entity type(s): agent, command, skill, mcp
2. task/domain keywords
3. source preference (GitHub slug, git URL, local path)
4. scope preference (local/global)
5. whether install should stay pinned (custom) or follow upstream (all)

2) Discover candidates

Run either aggregate or entity-specific find:

agentloom find <query>
agentloom agent find <query>
agentloom command find <query>
agentloom skill find <query>
agentloom mcp find <query>

3) Import the chosen resources

Use aggregate add for mixed imports:

agentloom add <source>

Use entity-scoped add when the user wants explicit selection:

agentloom agent add <source> --agents <csv>
agentloom command add <source> --commands <csv>
agentloom skill add <source> --skills <csv>
agentloom mcp add <source> --mcps <csv>

4) Create from scratch when needed

Create canonical files directly:

• Agent: .agents/agents/<name>.md
• Command: .agents/commands/<name>.md
• Skill: .agents/skills/<skill-name>/SKILL.md
• MCP: .agents/mcp.json

Agent template:

---
name: pr-reviewer
description: Reviews pull request diffs and reports concrete issues.
claude:
  model: sonnet
codex:
  model: gpt-5.3-codex
  reasoningEffort: low
  webSearch: true
---

You are a strict PR reviewer.
Focus on correctness, regressions, and missing tests.
Return findings with file references and concrete fixes.

Minimal command template:

Review the current branch diff and report bugs, regressions, and missing tests.

Minimal skill template:

---
name: release-check
description: Validate release readiness before publishing.
---

# Release Check

Run focused validation and report blockers with exact fix commands.

Minimal MCP template:

{
  "version": 1,
  "mcpServers": {
    "browser": {
      "base": {
        "command": "npx",
        "args": ["browser-tools-mcp"]
      }
    }
  }
}

5) Sync provider outputs

After create/import/update/delete, sync generated provider files:

agentloom sync

Optional provider targeting:

agentloom sync --providers cursor,claude,codex,opencode,gemini,copilot

6) Update and cleanup

Refresh tracked imports:

agentloom update [source]
agentloom agent update [source]
agentloom command update [source]
agentloom skill update [source]
agentloom mcp update [source]

Remove stale entries:

agentloom delete <source|name>
agentloom agent delete <source|name>
agentloom command delete <source|name>
agentloom skill delete <source|name>
agentloom mcp delete <source|name>
agentloom mcp server delete <name>

Response Contract

When using this skill, always provide:

1. exact commands to run
2. exact file path(s) to create/edit when authoring from scratch
3. sync command required after changes
4. concrete next step (find, import, create, update, sync, or delete)