Explain Yourself

Explain work so the user can inspect and act on it. Report outcomes, actions, evidence, assumptions, alternatives, and uncertainty without exposing or inventing hidden chain-of-thought.

Workflow

1. Select the explanation mode: plan, status, decision rationale, handoff, or postmortem. Read references/explanation-modes.md for the mode-specific contract.
2. Reconstruct from observable artifacts: the user's request, files, diffs, commands, tool results, tests, logs, source citations, and explicit decisions. Do not fill evidence gaps from plausible-sounding recollection.
3. Separate four layers:
   • Observed: directly supported facts and artifacts.
   • Reasoning summary: concise criteria and causal explanation connecting evidence to the decision.
   • Assumed: unstated conditions that affected the approach.
   • Uncertain: missing evidence, unresolved alternatives, risks, or confidence limits.
4. Explain proportionately. Lead with the outcome, include only actions that change interpretation, name rejected alternatives only when they clarify a tradeoff, and attach evidence near each material claim.
5. Validate the result against references/reasoning-safety.md. If asked for hidden reasoning, private scratch work, token-by-token thought, or chain-of-thought, decline that portion and provide a concise reasoning summary instead.
6. End with the current state, known gaps, and next action when the selected mode calls for them.

Output Rules