UNPKG

automagik-genie

Version:

Self-evolving AI agent orchestration framework with Model Context Protocol support

102 lines (77 loc) 2.76 kB
# Create Collective Agents Specialized agents for human-world work (non-coding). Each agent has persistent memory via session management. ## Core Agents (Always Present) ### researcher **Created:** Core agent (foundational) **Purpose:** Investigate topics, curate sources, synthesize findings **Use when:** Need evidence-backed research before creating content **Workflows:** Topic investigation, source validation, synthesis **Session pattern:** `researcher-<topic>` ### writer **Created:** Core agent (foundational) **Purpose:** Content creation from briefs and research **Use when:** Creating blog posts, documentation, marketing content **Workflows:** Content drafting, structure creation, voice consistency **Session pattern:** `writer-<content-type>` ### editor **Created:** Core agent (foundational) **Purpose:** Content refinement, polish, quality assurance **Use when:** Content needs review, editing, quality improvement **Workflows:** Copy editing, tone adjustment, clarity enhancement **Session pattern:** `editor-<content-id>` ### install **Created:** Core agent (foundational) **Purpose:** Setup, initialization, and onboarding **Use when:** Installing Genie, creating new workspaces, migrations **Workflows:** Project initialization, dependency setup, configuration **Session pattern:** `install-<context>` ### wish **Created:** Core agent (foundational) **Purpose:** Wish lifecycle management and orchestration **Use when:** Creating, tracking, and completing wishes **Workflows:** Wish creation, milestone tracking, completion validation **Session pattern:** `wish-<wish-id>` ## Emergent Agents (No emergent agents yet - agents emerge when patterns of repeated work appear) ## Agent Generation Philosophy Create doesn't come with pre-built agents for every scenario. Instead, agents **emerge** when: - User requests same type of work 3+ times - Complex domain requires persistent expertise - Multi-step workflow repeats regularly See `@.genie/create/spells/agent-generation.md` for complete generation protocol. ## Agent Architecture All agents follow this structure: ```markdown --- name: agent-name description: One-line specialty genie: executor: CLAUDE_CODE background: true --- # Agent Name • Identity & Mission [Purpose and expertise] ## Specialty [Unique capability] ## Operating Patterns [Workflows and approaches] ## Delegates To [Other agents this works with] ## Session Management [Session naming pattern] @AGENTS.md ``` ## Usage **Invoke agent directly:** ```bash genie run create/<agent-name> "<task description>" ``` **Let orchestrator route:** ```bash genie "<task description>" # Orchestrator determines appropriate agent ``` **Resume agent session:** ```bash genie resume <agent-name>-<context> ```