[FEATURE]: Direct Tool Access for Humans - Bridge the Agent-Human Capability Gap

[mlanza]

Feature hasn't been suggested before.

• [x] I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

Right now, there's a fundamental asymmetry in OpenCode: agents can execute custom tools and MCP tools, but humans cannot. When an agent struggles or when I want to demonstrate the right approach, I'm forced to either:

1. Type verbose instructions hoping the agent gets it right
2. Switch to manual shell commands, losing the safety and validation of custom tools
3. Watch the agent fumble through operations I could do in seconds

The Vision: Agent-Human Tool Parity

Give humans the same direct tool access that agents enjoy. Imagine typing:

$ tool search --pattern "TODO" --path "./src"
$ tool refactor --file "utils.ts" --type "extract-function"
$ tool test --coverage --threshold 80

Why This Resonates with Users

1. Mentorship Mode When agents get stuck, humans should be able to demonstrate the right sequence of operations. "Watch me do it once" is far more effective than describing it in words.
2. Rapid Prototyping Sometimes I just need to quickly validate an approach before involving the agent. Direct tool access lets me test hypotheses without the conversational overhead.
3. Debugging & Recovery When the agent goes off the rails, I need to manually intervene and get things back on track. Having the same tools means I can fix what broke and hand it back cleanly.
4. Learning & Discovery Humans exploring tools directly will discover powerful combinations they can then teach agents. This creates a virtuous cycle of capability expansion.
5. Trust & Transparency Being able to manually execute the same operations builds trust. I can verify tools work as expected before delegating to the agent.

The Technical Win

This leverages your existing custom tools system - no new security model needed. You've already built the safe, validated interface. Just expose it to humans.

User Impact

• Beginners: Learn by doing, not just watching
• Experts: Faster workflows, less friction
• Teams: Better collaboration, shared tool vocabulary
• Everyone: Reduced frustration when agents struggle

This isn't just about convenience - it's about making human-AI collaboration truly bidirectional. Sometimes the human should drive, and the tools shouldn't be locked behind an AI interface.

[github-actions[bot]]

This issue might be a duplicate of or related to existing feature requests. Please check:

• #5147: Plugin Support for Modal Human-in-the-loop Interactions - addresses human control and intervention in workflows
• #4355: OpenCode REPL mode - a simpler CLI interface that could complement direct tool access

Feel free to ignore if these don't address your specific case.