Multiple CLI Agents #83
Description
Problem Statement
Auto-Claude is designed as a desktop workflow for “autonomous multi-session AI coding” with a Kanban board, parallel agents, agent terminals, and git worktrees for safe isolation. It currently emphasizes “Run Claude Code in up to 12 terminals… with one-click task context injection” and parallel/self-validating flows.
In practice, teams often need multi-agent planning and execution:
Claude may be best for planning/architecture and high-context reasoning.
Codex CLI / Gemini CLI / Cursor agent / Grok may be better for:
quick fixes
repetitive refactors
boilerplate generation
fast iterations and cheaper runs
Today, Auto-Claude’s task execution model is effectively “Claude-only” (by design focus), which creates a gap:
No standardized way to assign tasks to different CLI agents from the same Kanban board
No governance to optimize cost vs complexity (simple tasks still consume expensive agent time)
No consistent mechanism to normalize outputs (diffs/logs/status) across different agent CLIs
No “agent capability routing” (which agent should pick this card?)
Auto-Claude already provides the right workflow shell; it needs an agent abstraction layer to support multiple agent runtimes while preserving the existing Kanban-driven workflow.
Proposed Solution
Extend Auto-Claude so that each Kanban task can be executed by a selected agent provider (Claude Code, Codex CLI, Gemini CLI, etc.)
Alternatives Considered
No response
Component
Python Backend (auto-claude/)
How important is this feature to you?
None
Contribution
- I'm willing to submit a PR for this feature
Additional Context
No response