Review Tasks
Use --no-commit to run the agent in analysis or review mode without letting it modify git history.
When to use this
- Code review: "what are the potential security issues in this module?"
- Architecture analysis: "explain how the auth flow works"
- Test planning: "what edge cases should we test for the payment processor?"
- Any task where you want agent output but don't want code changes committed
Usage
CLI:
bash
ysa run "review the error handling in src/api and list what's missing" --no-commitAPI:
ts
await runTask({
...
allowCommit: false,
});What changes
With --no-commit / allowCommit: false:
- The
Bashtool is still available — the agent can read files, run tests, etc. - The agent cannot run
git commit,git push, or related commands (the tool is removed from its allowed set) - The worktree still exists after the task — you can inspect any files the agent created or modified
Combining with quiet mode
For analysis tasks where you only care about the final answer:
bash
ysa run "summarize the database schema" --no-commit --quiet--quiet suppresses the streaming output and only shows progress markers, so the final agent message is the first thing you read after the task completes.
Reviewing the output
bash
ysa logs <task-id> | jq 'select(.type == "assistant") | .text'Or just:
bash
ysa logs <task-id>Related
ysa run— full CLI reference- RunConfig.allowCommit — API reference