{
  "markdown": "# claude-plugins\n\nPersonal Claude Code plugin — 14 specialized agents with an orchestrator-first workflow.\n\n## Usage\n\n### Load for a single session\n\n```bash\nclaude --plugin-dir ./claude-plugins\n```\n\n### Install globally (user scope)\n\nRequires a marketplace. See [Plugin marketplaces](https://code.claude.com/docs/en/plugin-marketplaces) to self-host one, then:\n\n```bash\nclaude plugin install plugins@fahimtran\n```\n\n### Reload after changes\n\n```\n/reload-plugins\n```\n\n## How it works\n\nWhen the plugin is active, `settings.json` sets `orchestrator` as the main agent. The orchestrator breaks down tasks and delegates to the appropriate specialized agent. Simple questions are answered directly without delegation.\n\n## Agents\n\n| Agent | Model | Role |\n|---|---|---|\n| **orchestrator** | inherited | Workflow coordinator — delegates to other agents |\n| **architect** | inherited | System design and implementation planning |\n| **ask** | inherited | Read-only Q&A and explanations |\n| **code** | sonnet | Hands-on implementation (default workhorse) |\n| **code-reviewer** | sonnet | Pre-commit code review |\n| **db** | sonnet | Database schema, migrations, queries |\n| **debug** | sonnet | Bug investigation and root cause analysis |\n| **devops** | sonnet | Docker, CI/CD, infrastructure |\n| **docs** | sonnet | Documentation writer |\n| **pr-writer** | inherited | PR titles and descriptions |\n| **refactor** | sonnet | Code restructuring without behavior changes |\n| **researcher** | inherited | External research via web search |\n| **security** | inherited | Security auditing (read-only) |\n| **tester** | sonnet | Test writing |\n\n\"Inherited\" means the agent uses whatever model is active in your Claude Code session.\n\n### Design principles\n\n- **Orchestrator-first**: `settings.json` activates the orchestrator as the main thread for all non-trivial tasks.\n- **Parallel execution**: The orchestrator spawns independent subtasks (e.g. multiple file writes) as parallel agents rather than sequentially.\n- **Model tiering**: Planning/research agents inherit the session model; implementation agents use sonnet for speed.\n- **Permission modes**: Read-only agents (`ask`, `architect`, `code-reviewer`, `pr-writer`, `researcher`, `security`) use `permissionMode: plan` to prevent accidental writes.\n- **Clear boundaries**: Each agent has explicit \"What You Should NOT Do\" guardrails.\n\n## Structure\n\n```\nclaude-plugins/\n├── .claude-plugin/\n│   └── plugin.json       # Plugin manifest\n├── agents/               # 14 agent definitions\n├── settings.json         # Activates orchestrator as main agent\n└── README.md\n```\n\n## Requirements\n\n- Claude Code >= 1.0.33\n",
  "bytes": 2646,
  "sha": "48e619a36f09d8e843d23b9bcdcb6f949cf237be4a9756fd980f35cf8f1b1299",
  "repo_slug": "fahimtran/claude-plugins",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/plg_fahimtran_claude_plugins_fahimtran_5e6a7481/readme"
}