← ClaudeAtlas

perfecting-presentationslisted

Use when the user wants to review, audit, polish, or perfect an existing HTML slide deck until it is genuinely done, for example "review this deck until it is perfect", "make this presentation crisp", "audit my slides". Runs a strict convergence loop, deterministic hard gates plus fresh-context judge panels with adversarial verification, terminating only when consecutive review rounds come back dry. For building a deck see workbench:crafting-presentations; for designing slide content see writing:presentations.
pgoell/pgoell-claude-tools · ★ 0 · Web & Frontend · score 70
Install: claude install-skill pgoell/pgoell-claude-tools
# Perfecting Presentations A strict reviewer for HTML decks that does not stop at "looks good". The deck is done when fresh reviewers can no longer find anything that survives adversarial scrutiny against an explicit standard. This skill defines that standard (the constitution), the deterministic checks (hard gates), the review loop (judge, verify, fix), and the termination rule (dry rounds). ## When to use this skill Reach for `perfecting-presentations` when a deck already exists and the user wants it reviewed, audited, polished, or "perfected". It pairs with `workbench:crafting-presentations` decks (a `<deck-stage>` element with `<section>` slides) but the loop works for any HTML deck once you identify the slide container. It is not for designing content (use `writing:presentations`) or for first-pass rendering (use `workbench:crafting-presentations`). ## Perfection as convergence A single reviewer verdict is worthless as a termination condition: a self-graded loop collapses into early self-approval, and a "be very strict" reviewer never terminates and oscillates (round 3 undoes round 2). So this skill never asks "is the deck perfect?". It asks "did this round find anything new that survives verification?" and stops only when: 1. every hard gate passes, and 2. two consecutive review rounds produce zero verified findings at or above the severity threshold, and 3. the round cap (default 5) has not been hit. If it is hit, stop and report the remaining open findings hones