How do you make sure everything is really done-done

Claude agent double-check prompt

Pheromind Discord Tip

Perform a rigorous, exhaustive validation of all implemented features against the original intent. Explicitly identify and detail any:

  • Missing components or unimplemented features.,
  • Functional defects or broken aspects.,
  • Unintended behaviors or elements that work but shouldn’t.,
  • Falsely reported “completed” features that do not yet function.,

Provide a self-assessment score (0-100) that accurately reflects the true completeness and correctness of the work. If your score is less than 100, provide a granular explanation for each point deducted. Strategically orchestrate parallel and sequential sub-agents to independently address and resolve every identified issue. Ensure seamless context transfer and strict task isolation for each sub-agent. Following the completion of each sub-agent’s task, immediately deploy a dedicated review sub-agent to verify the fix and implement further corrections until the issue is entirely resolved and the work is verifiably 100% complete and correct. The final output must represent a state where all aspects of the initial request are fully implemented, functional, and accurately represented, with no “not yet implemented” features persisting or obscured.

Notes (2026-09-30)

  • The prompt is a community tip attributed to the “Pheromind” Discord; no primary source was found or fetched, so provenance is unverified. It is a prompt template, not a verifiable claim set.
  • It assumes Claude Code sub-agents; see claude-code-agents-convention for current subagent mechanics (subagents can be spawned in parallel, default nesting depth 3).
  • A self-reported 0-100 score is an LLM self-assessment, not a measurement; pair with real tests and review.

Sources