Agent Skills
› mohitagw15856/pm-claude-skills
› deck-review-rubric
deck-review-rubric
GitHub基于论点、证据、密度、叙事弧和诚实度五维标准，对演示文稿进行结构化评审。提供分级反馈与核心修改建议，避免主观审美干扰，确保反馈可执行且聚焦关键问题。
Trigger Scenarios
review my deck
give feedback on this presentation
is this ready for the board
our deck reviews are just font opinions
Install
npx skills add mohitagw15856/pm-claude-skills --skill deck-review-rubric -g -y
SKILL.md
Frontmatter
{
"name": "deck-review-rubric",
"description": "Review a deck against a rubric instead of taste — the five dimensions (argument, evidence, density, arc, honesty), the severity-sorted feedback that separates broken from suboptimal, and the review conversation that improves the deck without rewriting it in the reviewer's voice. Use when asked review my deck, give feedback on this presentation, is this ready for the board, or our deck reviews are just font opinions. Produces the rubric scores with evidence, the severity-sorted feedback, the two-fixes-that-matter-most call, and the reviewer discipline notes."
}
Deck Review Rubric Skill
Deck reviews default to taste: font opinions, color takes, and "I'd reorder this" — the reviewer's aesthetic imposed at slide grain while the deck's actual failure (no argument, no ask, dishonest chart) sails through unexamined. The rubric reviews what decks fail at: argument (do the headlines make a case — deck-outline-first skim test), evidence (does each slide prove its claim — evidence-grading weight), density (billboard or wall — slide-density-rules), arc (does tension carry the room — deck-narrative-arc), and honesty (axes, claims, caveats — citation-hygiene). Findings sort by severity — broken beats suboptimal beats taste — and the review ends with the only sentence presenters can act on: the two fixes that matter most.
What This Skill Produces
- The rubric scores — the five dimensions, each with the evidence for its score (cited slides, not vibes)
- The severity-sorted findings — 🔴 broken (would fail in the room) / 🟡 suboptimal (costs force) / ⚪ taste (offered, labeled)
- The two-fixes call — the highest-leverage changes, because forty comments produce paralysis and two produce a better deck
- The reviewer discipline — the review-comments-resolver contract honored from the giving side
Required Inputs
Ask for these if not provided:
- The deck and its job — audience, ask, time slot; a deck can't be reviewed against an unknown mission (the most common finding is that the mission itself is undefined — that's a 🔴, not a formatting note)
- The review's timing — outline stage (structure feedback is cheap — the ideal), draft (full rubric), night-before (triage: 🔴 only, kindly)
- The presenter's stakes and experience — a first-time board presenter needs the confidence-preserving version: same findings, framed as upgrades
- What the presenter wants checked — their stated worry gets explicit attention (and the review says what it found there)
Framework: The Rubric Rules
- Argument first — the skim test: read only the headlines: is there a case with an ask? Topic-titled slides, missing asks, and sequences that don't argue are 🔴 findings that outrank everything visual — a beautiful deck with no argument is a screensaver.
- Evidence per claim: each headline's slide either proves it or doesn't — overclaimed charts, anecdotes carrying prevalence claims, and the missing caveat are the substance findings. The honesty sweep rides here: axes, windows, sources, the "studies show" ban.
- Severity discipline is the review's value: 🔴 = fails in the room (broken argument, dishonest chart, no ask) · 🟡 = costs force (walls of text, buried wow, sagging middle) · ⚪ = taste (labeled as such, take-or-leave). A review that presents font kerning and a missing ask at equal volume has abdicated its job.
- Two fixes, named: the closing call — "fix these two and this deck works: (1) the ask is missing, add slide 9; (2) slides 4–6 are one claim, merge them" — because the presenter has one evening, and the rubric's forty observations compress into the two that move the outcome.
- Review the deck, not your deck: the reviewer's restructuring instinct ("I'd tell this differently") is ⚪ unless the current structure fails the rubric — the deck stays the presenter's (house-style-enforcer voice-preservation logic). Timing shapes kindness: night-before reviews deliver 🔴s only, with the fix attached, and save the 🟡 education for the retro.
Output Format
Deck Review: [deck] — for [audience/ask] · stage: [outline/draft/final]
Rubric Scores
| Dimension | Score | Evidence (slides cited) |
|---|---|---|
| [Argument · Evidence · Density · Arc · Honesty] |
Findings (severity-sorted)
🔴 Broken: [each: slide, what fails in the room, the fix] 🟡 Costs force: […] ⚪ Taste (labeled): […]
The Two Fixes
[1 · 2 — the highest-leverage pair, with why]
On Your Stated Worry
[What the review found about the thing they asked about]
Quality Checks
- All five dimensions scored with slide citations
- Every finding carries its severity honestly — no taste smuggled into 🟡
- The two-fixes call exists and is genuinely highest-leverage
- The presenter's stated worry got an explicit answer
- Night-before reviews delivered 🔴 only
Anti-Patterns
- Do not review taste at finding volume — label it ⚪ or leave it home
- Do not equalize severities — the missing ask and the font choice are different emergencies
- Do not rewrite the deck in your voice — rubric failures get fixes; style differences get silence
- Do not deliver forty comments — the two-fixes compression is the review's actual deliverable
- Do not soften 🔴s for kindness — the room will deliver them unsoftened tomorrow
Version History
- 471c606 Current 2026-07-22 14:47
