training-slide-reviewer - SKILL.md Agent Skill

name: training-slide-reviewer description: > Evaluate slides for narrator mode, working-memory gaps, and inferred expectation signals. Two-axis review plus expectation audit. Use for "review my slides", "check my deck", "slide review". version: 2.0.0

Slide Reviewer

Context: Tim's Slide Relationship

Tim has low working memory and compensates by leaning on slides. Two failure modes:

Slide Narrator Mode — Slides carry so much content Tim reads them. Energy dies.
Working Memory Gap — Slides too sparse, Tim loses his thread.

Sweet spot: slides that ANCHOR Tim's delivery without REPLACING it.

Core Behavior

Pull slide deck from Canva (via connector) or accept uploaded images/PDF
Map each slide to the session run-of-show (if available)
Evaluate each slide on two axes plus expectation signals

Axis 1: Content Load (is the slide doing too much?)

RED: Full sentences Tim will read verbatim. "This slide is a teleprompter."
YELLOW: Key points present but could be condensed. "Trim to keywords."
GREEN: Keywords, visuals, or framework only. "Anchors without narrating."

Axis 2: Anchor Strength (does it support Tim's working memory?)

RED: Blank or decorative only. Tim will lose his place.
YELLOW: Some structure but missing concept label.
GREEN: Clear concept anchor with visual support.

Axis 3: Expectation Signals (NEW — what does this slide tell the audience to expect?)

Slides create INFERRED expectations. The audience reads a slide and forms an expectation about what the session IS, what's coming next, and what kind of experience they're having.

RED: Slide promises something the session doesn't deliver. (e.g., slide says "Workshop Exercises" but the session is all lecture; slide lists 8 topics but only 3 get covered)
YELLOW: Slide is ambiguous — audience could interpret it as a promise. (e.g., agenda slide with items that may get cut; slide with "Q&A" that might not happen)
GREEN: Slide accurately represents what's coming. No false promises.

Output Format

SLIDE REVIEW: [Session Title] — [N slides for X min session]

SLIDE RATIO: [N slides / X min = Y slides per minute]
TARGET: ~1 slide per 2-3 min for keynote, ~1 per 1-2 min for workshop
VERDICT: [Too many slides / Just right / Too few]

SLIDE | CONTENT | ANCHOR | EXPECT. | ISSUE                          | FIX
──────|─────────|────────|─────────|────────────────────────────────|──────
  1   | GREEN   | GREEN  | GREEN   | —                              | —
  2   | RED     | GREEN  | GREEN   | Full paragraph on slide        | Cut to 3 keywords
  3   | GREEN   | RED    | GREEN   | Decorative, no anchor          | Add concept header
  5   | GREEN   | GREEN  | RED     | Lists 6 outcomes, only 3 get   | Remove outcomes
      |         |        |         | covered — creates broken promise| you won't deliver
  7   | YELLOW  | YELLOW | YELLOW  | 5 bullets, agenda might shift  | Pick top 2
 12   | RED     | GREEN  | GREEN   | Tim will read this verbatim    | Replace with visual
...   |         |        |         |                                |

SUMMARY:
- RED content load: [N] — Tim will narrate these.
- RED anchor: [N] — Tim will lose his place.
- RED expectation: [N] — Slides promise what session doesn't deliver.
- Slide-narrator risk: [HIGH/MEDIUM/LOW]
- Working-memory risk: [HIGH/MEDIUM/LOW]
- False-promise risk: [HIGH/MEDIUM/LOW]

Evaluation Rules

Slide titles should be headlines, not labels.
One idea per slide.
Tim's verbal skills carry detail. Slides carry structure.
Flag any slide that takes more than 3 seconds to read.
If 40+ slides for 45 minutes, flag the ratio immediately.
Agenda slides are expectation contracts. If the agenda lists something, the session must deliver it. Flag any agenda item that might get cut.
Title slides set the first inferred expectation. If the title says "Workshop" but it's a lecture, flag it.