Skip to content

Conversation

@dataforxyz
Copy link
Contributor

Summary

  • Add -o flag to override AI agent tool and/or model for specific workflow steps
  • Allows using different agents for different steps in a single task

Usage

rover task -o implement:claude:opus -o context:gemini "task description"

Changes

  • packages/cli/src/utils/step-agent-parser.ts - Parse "step:tool[:model]" format
  • packages/cli/src/program.ts - Add -o/--override flag
  • packages/cli/src/commands/task.ts - Parse and validate step overrides
  • packages/schemas/src/task-description/ - Add stepAgents field
  • packages/cli/src/lib/sandbox/ - Pass --step-agents to rover-agent
  • packages/agent/src/cli.ts - Add --step-agents option
  • packages/agent/src/lib/runner.ts - Use step-specific tool/model overrides

Test plan

  • Run rover task -o implement:gemini "test" and verify implement step uses gemini
  • Run rover task -o context:claude:opus -o implement:gemini:flash "test" with multiple overrides
  • Verify invalid step IDs are rejected with helpful error message

🤖 Generated with Claude Code

Add support for overriding the AI agent tool and/or model for specific
workflow steps using the -o flag at task creation time.

Usage:
  rover task -o implement:claude:opus -o context:gemini "task description"

This allows users to:
- Use different AI agents for different workflow steps
- Specify different models for specific steps
- Override workflow defaults on a per-task basis

Changes:
- Add step-agent-parser.ts for parsing "step:tool[:model]" format
- Add stepAgents field to task description schema
- Add -o/--override flag to task command in CLI
- Update sandbox to pass --step-agents to rover-agent
- Add --step-agents option to rover-agent run command
- Update Runner to use step-specific tool/model overrides

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants