Rocky Production Readiness Gates

Rocky is most useful when it can act, verify, and report without improvising around missing setup. This checklist gives teams a public-safe way to decide whether an AI operator is ready for real workflows.

Official Hermes Agent docs: https://hermes-agent.nousresearch.com/docs

Readiness map

• A working model/provider configuration is selected for the task class.
• Required toolsets are enabled before the run begins, not discovered halfway through the work.
• Approval mode is understood: manual for sensitive work, smart or off only when the operator boundary is trusted.
• Project files, public/private boundaries, and deploy commands are documented before mutation.
• A verification path exists for every promised artifact.

Daily operator checklist

• Run a small status or doctor check when configuration has changed.
• Confirm the target working directory and profile before editing files.
• Inspect source-of-truth files instead of relying on memory.
• Prefer official docs for commands and feature behavior.
• Use a narrow first action, then widen only after it succeeds.

Pitfalls

• Starting a production task before the deployment or rollback path is known.
• Treating a model response as evidence when a tool check is available.
• Letting the operator see private data when a public-safe fixture would work.
• Skipping mobile or live-route QA for public surfaces.

Verification steps

• Record the commands or URLs checked.
• Confirm the output still matches the user-visible goal.
• Check logs or screenshots when the task affects a public page.
• Summarize blockers honestly instead of inventing a successful result.