agentsclimarketplace

Adversarial quality loop

Skill twk0672005/sentence-to-site/skills/adversarial-quality-loop

One sentence in, a verified website out. Agent skills that close the gap between vibe coding and software engineering - so people with no coding background can build a website that holds up.

Install
npx -y skills add twk0672005/sentence-to-site --skill adversarial-quality-loop

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • 19 days oldThe repository was created 19 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
  • 5 stars5 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Use when a user explicitly requests an independent review, challenge pass, or second opinion on a rendered website or animation artifact. Produces one bounded evaluator report and evidence verdict; it never auto-starts another build pass.

SKILL.md

1.9 KB, as published. Nobody here has run it

Bounded Independent Review

This legacy-named skill is not a default quality loop. It is an optional, one-time independent review after a route has produced the required evidence.

Preconditions

  • The user explicitly requested independent review, a challenge pass, or a second opinion.
  • Acceptance criteria and current rendered evidence are available.
  • Separate reviewers are available. If not, report that independent review is unavailable; do not simulate it as a solo role-play.

Run one review pass

  1. Give an independent Evaluator the acceptance criteria and rendered artifact only.
  2. Give an independent Evidence Auditor the selected route's evidence requirements and artifacts.
  3. Return the Evaluator's finite report plus the Auditor's pass, partial, or blocked verdict.
  4. Offer the user's choices: accept, reject, defer, or start a new explicitly scoped implementation task.

The Evaluator may identify one visible concern and one recommendation. Neither reviewer edits code, asks for another review round, or authorizes another build pass.

Stop condition

Stop after the report and verdict. A new implementation pass begins only through a new user decision, not because an agent wants to improve its own output.

Boundaries

  • Self-critique does not count as independent review.
  • A missing reviewer is unavailable, not a reason to silently downgrade to solo review.
  • A partial or blocked evidence verdict propagates to delivery; it does not trigger retry automation.
  • See ../../references/quality-loop-roles.md for the bounded evaluator and auditor contracts.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.