From infinite-improbability-drive
Spawn a sandboxed LLM with multi-LLM coordination for automated code review
How this skill is triggered — by the user, by Claude, or both
Slash command
/infinite-improbability-drive:spawn-teamThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Launches an isolated LLM instance with coordinated review from a secondary LLM. This enables automated code review workflows where one LLM implements changes and another reviews them.
Launches an isolated LLM instance with coordinated review from a secondary LLM. This enables automated code review workflows where one LLM implements changes and another reviews them.
When you need delegated task execution with automated code review:
/spawn-team "implement the new authentication flow"
With coordination mode:
/spawn-team --sequential "straightforward feature"
/spawn-team --ping-pong "complex refactoring"
With custom LLMs:
/spawn-team --reviewer claude-code "use Claude as reviewer"
/spawn-team --primary gemini-cli --reviewer claude-code "swap roles"
Primary LLM completes the task, reviewer provides feedback, primary applies fixes once. Best for straightforward tasks.
Primary → Reviewer → Primary (fixes) → PR
Primary and reviewer alternate until the reviewer approves or max iterations reached. Best for complex tasks requiring iteration.
Primary → Reviewer → Primary → Reviewer → ... → PR
Control iterations:
/spawn-team --ping-pong --max-iterations 5 "complex task"
| Identifier | Description |
|---|---|
claude-code | Claude Code CLI (default primary) |
gemini-cli | Gemini CLI (default reviewer) |
Settings in .infinite-probability/improbability-drive.toml:
[spawn-team]
coordination = "sequential" # or "ping-pong"
max_iterations = 3
reviewer_llm = "gemini-cli"
[spawn]
default_llm = "claude-code"
The reviewer LLM receives:
Review response format:
{
"verdict": "approved",
"suggestions": []
}
Or with suggestions:
{
"verdict": "needs_changes",
"suggestions": [
{
"file": "src/auth.rs",
"line": 42,
"issue": "Missing error handling",
"suggestion": "Add Result return type"
}
]
}
When the reviewer returns needs_changes, the primary LLM receives:
Returns a SpawnTeamResult containing:
success: Whether the team operation succeedediterations: Number of review iterations performedfinal_verdict: Approved, NeedsChanges, or Failedreviews: All review results with suggestionssummary: Human-readable summaryUse spawn-team when:
Use basic spawn when:
Same isolation guarantees as /spawn:
$HOME or config files--dangerously-skip-permissions never allowedUser: /spawn-team --ping-pong "add rate limiting to the API"
Spawn-team:
1. Primary (claude-code) implements rate limiting
2. Reviewer (gemini-cli) finds edge case: "No handling for burst traffic"
3. Primary addresses feedback, adds burst handling
4. Reviewer approves changes
5. PR created with 2 review iterations documented
Result:
- Status: Success
- Iterations: 2
- Final verdict: Approved
- PR: https://github.com/repo/pull/123
npx claudepluginhub epiphytic/infinite-improbability-drive-v0Runs cross-LLM iterative code reviews with Codex or Gemini CLI peers, applying accepted fixes until consensus on improved code and report.
Orchestrates teams of Claude Code sessions for parallel work on research, features, debugging, and cross-layer tasks using shared tasks, inter-agent messaging, and team lead coordination.
Coordinates multiple Claude Code instances as agent teams with shared tasks, inter-agent messaging, and management for parallel research, feature development, debugging, and cross-layer coordination.