What it does
validate-distilled-coding-standards runs a controlled experiment on your distilled standards: for each historical review comment, one isolated subagent re-reviews the same code armed with only the standards — never the original comment, the PR discussion, or the current conversation. Agreement between agent and human proves the standards transmit review intent on their own; every miss is an edit the standards need, and the misses are the experiment’s real product.
It is deliberately expensive — roughly 40–50k tokens per example — so it warns you with a cost estimate and waits for an explicit yes before fanning out.
When to reach for it
- Invocation mode. You invoke this by typing
/validate-distilled-coding-standards— the agent won’t reach for it on its own. - Trigger boundary. Reach for it after a distill-coding-standards run, when you want proof the standards actually carry your review voice — not for reviewing new code, which is apply-coding-standards’ job.
Prerequisites
The dossier left by /distill-coding-standards at .claude/coding-standards-dossier/, plus the standards docs it produced. Without a dossier it can rebuild just the slice the lab needs, at extra cost.
The lab
Each example is a clean code excerpt paired (invisibly, to the agent) with the human comment it originally drew. Verdicts are three-valued — matched, matched and went further, or different angle — with honesty rules: an artifact of the excerpt being cut mid-element counts as neither hit nor miss, and a defensible deviation is a different angle, not a failure. The output is a published lab page plus two concrete miss-lists: rules that exist but were under-applied, and violations no rule covers — ready to apply as edits on your word.
It’s working if
- You were told the example count and token estimate, and said yes, before anything ran.
- Every example on the lab page shows the code, the human comment, the agent’s problems, and a verdict.
- The run ends with proposed standards edits, not just scores.
Where it fits
The verification step of the coding-standards chain — periodic maintenance after a distillation, not a per-commit tool. distill-coding-standards produces what it tests, because both share the dossier; apply-coding-standards is the consumer that benefits from every miss it fixes. For the map of the whole set, see ask-eddy.