Run OpenAI Codex CLI agent (deep analysis / critical review).

NO SHARED MEMORY:

• Cannot see messages/outputs from gemini/claude/opencode.

• Only sees: (1) this prompt, (2) files in context_paths, (3) its own history via continuation_id.

CROSS-AGENT HANDOFF:

• Small data: paste into prompt.

• Large data: save_file -> context_paths -> prompt says "Read ".

CAPABILITIES:

• Strongest deep analysis and reflection abilities

• Excellent at finding issues, edge cases, and potential bugs

• Good at critical code review and architectural assessment

BEST PRACTICES:

• Be explicit about scope: "Only fix X, don't refactor Y"

• Specify constraints: "Keep it simple, no new abstractions"

Supports: image attachments.

Run multiple codex tasks in parallel. All tasks share workspace/permission/save_file. Results are appended to save_file with XML wrappers (). Max 100 tasks. Model can be array: single element shared by all, or one per task.

Run Google Gemini CLI agent (UI design / comprehensive analysis).

NO SHARED MEMORY:

• Cannot see messages/outputs from codex/claude/opencode.

• Only sees: (1) this prompt, (2) files in context_paths, (3) its own history via continuation_id.

CROSS-AGENT HANDOFF:

• Small data: paste into prompt.

• Large data: save_file -> context_paths -> prompt says "Read ".

CAPABILITIES:

• Strongest UI design and image understanding abilities

• Excellent at rapid UI prototyping and visual tasks

• Great at inferring original requirements from code clues

• Best for full-text analysis and detective work

BEST PRACTICES:

• Good first choice for "understand this codebase" tasks

Run multiple gemini tasks in parallel. All tasks share workspace/permission/save_file. Results are appended to save_file with XML wrappers (). Max 100 tasks. Model can be array: single element shared by all, or one per task.

Run Anthropic Claude CLI agent (code implementation).

NO SHARED MEMORY:

• Cannot see messages/outputs from codex/gemini/opencode.

• Only sees: (1) this prompt, (2) files in context_paths, (3) its own history via continuation_id.

CROSS-AGENT HANDOFF:

• Small data: paste into prompt.

• Large data: save_file -> context_paths -> prompt says "Read ".

CAPABILITIES:

• Strongest code writing and implementation abilities

• Excellent at translating requirements into working code

• Good at following patterns and conventions

BEST PRACTICES:

• Be explicit about target: "Replace old implementation completely"

• Specify cleanup: "Remove deprecated code paths"

Supports: system_prompt, append_system_prompt, agent parameter.

Run multiple claude tasks in parallel. All tasks share workspace/permission/save_file. Results are appended to save_file with XML wrappers (). Max 100 tasks. Model can be array: single element shared by all, or one per task.

Run OpenCode CLI agent (full-stack development).

NO SHARED MEMORY:

• Cannot see messages/outputs from codex/gemini/claude.

• Only sees: (1) this prompt, (2) files in context_paths, (3) its own history via continuation_id.

CROSS-AGENT HANDOFF:

• Small data: paste into prompt.

• Large data: save_file -> context_paths -> prompt says "Read ".

CAPABILITIES:

• Excellent at rapid prototyping and development tasks

• Good at working with multiple frameworks and tools

• Supports multiple AI providers (Anthropic, OpenAI, Google, etc.)

BEST PRACTICES:

• Specify agent type for specialized tasks (e.g., --agent build)

• Use file attachments for context-heavy tasks

Supports: file attachments, multiple agents (build, plan, etc.).

Run multiple opencode tasks in parallel. All tasks share workspace/permission/save_file. Results are appended to save_file with XML wrappers (). Max 100 tasks. Model can be array: single element shared by all, or one per task.

Generate images using Nano Banana Pro (Gemini 3 Pro Image).

CAPABILITIES:

• Text-to-image generation with high quality output

• Image editing and transformation with reference images

• Multiple aspect ratios and resolutions (1K/2K/4K)

• Style transfer and multi-image fusion

• Optional search grounding for factual content

RESPONSE FORMAT:

• Returns XML with file paths to generated images

• Images are saved to disk (no base64 in response)

• Includes text descriptions and optional thinking process

BEST PRACTICES:

• Be descriptive: describe scenes, not just keywords

• Use negative constraints in prompt: "no text", "no watermark"

• For editing: provide reference image and specify what to keep

• For style transfer: provide style reference image

Supports: reference images with roles (edit_base, style_ref, etc.).

Generate images via OpenRouter-compatible or OpenAI-compatible endpoints.

CAPABILITIES:

• Text-to-image generation with multiple providers

• Image editing and transformation with reference images

• Multiple aspect ratios and resolutions (1K/2K/4K)

RESPONSE FORMAT:

• Returns XML with file paths to generated images

• Images saved to disk (no base64 in response)

• Includes text descriptions when available

BEST PRACTICES:

• Be descriptive: describe scenes, lighting, style, composition

• Use negative constraints in prompt: "no text", "no watermark", "no blur"

• For editing: provide reference image and specify what to keep

Supports: reference images for editing.

Get the GUI dashboard URL. Returns the HTTP URL where the live event viewer is accessible.