gemini-cli - 💡(How to fix) Fix Docs guidance: auditing final agent output claims?

Summary

Would the project be open to a small docs-only note about auditing final agent output claims in non-interactive or automation workflows?

The idea is not to introduce a new dependency or change Gemini CLI behavior. It would be a compact safety guidance pattern for prompts and downstream scripts that consume final CLI output.

Motivation

Gemini CLI supports terminal-first automation, GitHub workflows, and structured output modes such as --output-format json and --output-format stream-json.

In those contexts, a final answer may contain operational claims such as:

• tests passed
• files were changed
• a PR is ready for review
• a workflow is ready to deploy or publish
• a human approval is no longer needed

Those claims may need evidence from command output, changed files, CI logs, structured events, or human approval before downstream automation treats the task as complete.

Possible docs shape

A short section could suggest asking the agent to include a final receipt with:

verification_status: missing_evidence
summary: "The agent claimed tests passed, but no test command output was provided."
claims:
  - claim: "All tests passed."
    support_status: unsupported
    evidence: []
    required_fix: "Attach the test command and output, or downgrade the claim."
next_owner: reviewer
human_decision_required: false

This would complement existing output modes and review/CI guidance by making final-state claims easier for humans and scripts to inspect.

Question

If this fits the project, where would maintainers prefer this docs-only guidance to live?

Potential locations might be the automation docs, CLI output-format docs, or a safety/security-oriented docs page.