Summary
Add a run_command tool that allows the model to execute shell commands in the jam run agentic loop.
Details
The file src/tools/run_command.ts already exists but may need to be registered and integrated into the tool registry.
Requirements
- Ensure
run_command tool is properly implemented and registered
- Must be a write tool (requires user confirmation before execution)
- Should have safety guards: no
rm -rf, no sudo, configurable via allowlist
- Capture stdout and stderr, return to the model
- Timeout after configurable duration (default: 30s)
Safety Considerations
- This is the highest-risk tool — confirmation prompt must clearly show the command
- Consider a separate
dangerousCommandPatterns config option
Acceptance Criteria
Summary
Add a
run_commandtool that allows the model to execute shell commands in thejam runagentic loop.Details
The file
src/tools/run_command.tsalready exists but may need to be registered and integrated into the tool registry.Requirements
run_commandtool is properly implemented and registeredrm -rf, nosudo, configurable via allowlistSafety Considerations
dangerousCommandPatternsconfig optionAcceptance Criteria