| name | autopilot-predict |
| description | Use learned patterns and current state to predict the optimal next Codex action |
| argument-hint | |
| allowed-tools | mcp__claude-flow__autopilot_predict mcp__claude-flow__autopilot_progress mcp__claude-flow__autopilot_learn mcp__claude-flow__autopilot_history |
Predict what to work on next using Ruflo autopilot intelligence:
- Call
mcp__claude-flow__autopilot_predict for the recommended next action.
- If confidence is above 0.7, execute the prediction directly in Codex.
- If confidence is below 0.7, check
mcp__claude-flow__autopilot_progress for task breakdown.
- Pick the highest-priority incomplete task that is safe to handle in the current Codex turn.
- After completing work, call
mcp__claude-flow__autopilot_learn to update patterns.
Learning Pipeline
mcp__claude-flow__autopilot_learn -- discover success patterns from completed tasks.
mcp__claude-flow__autopilot_history({ query: "KEYWORD" }) -- search past completions.
- Patterns are stored in AgentDB for cross-session recall.
Codex Iteration Model
Codex does not provide Claude Code /loop semantics. Use this skill as a bounded next-action selector:
- High confidence prediction -> execute immediately when it fits the user's current request.
- Low confidence -> fall back to explicit task priority or ask for clarification if the next action is risky.
- No tasks remaining -> disable autopilot or report that no further work is available.
For repeated unattended execution, use an external scheduler or a user-requested follow-up session rather than assuming Codex can wake itself.