PPTX highest priority

Video-only talks are processable

No-video talks not processable

Multiple slide_source values used

Video extraction output named by YouTube ID

Slides fail + video available = fallback to video

Partial processing when one source fails

All fail = distinct failure status

Processing log explains decisions

Regenerate for content additions

Edit for content removal

Simplified anchor flagged

Safety suffix: no new elements

Safety suffix: background continuation

Explicit preservation instructions

Versioned output naming

Fix mode for near-perfect images

PIL masking flagged as wrong approach

Missing preservation flagged

Content modification via edit flagged

Opening pattern analyzed

Narrative structure analyzed

Humor documented

Audience interaction noted

Areas for improvement

Structured data fields

Opening type categorized correctly

Closing type categorized correctly

Verbatim signature phrases

Verbatim opening lines

Verbatim closing lines

Transition phrases extracted

JSON schema completeness

Required top-level keys

Speaker fields from config

Speaker badges present

Speaker badges grounded in vault stats

Signature patterns with usage counts

never_used_patterns listed

guardrail_sources.slide_budgets is array

instrument_catalog populated

confirmed_intents carried over

pacing data populated

antipattern_frequency present

Analysis file location

All 14 dimensions present

English-only quote format

Language-tagged verbal signatures

Structured data fields

Pattern score formula

Observable patterns only

Verbatim examples

Tracking DB updated

Areas for improvement present

Presentation Patterns Scoring section

Abstract word count

Takeaway count

Speaker bio included

Title present

Presentation spec produced

Spec includes mode

No full architecture

Speaker voice in abstract

Spec includes duration

Spec includes profanity register

Spec includes commercial intent

Spec includes shownotes slug

Anti-sell awareness

Per-joke grading

Joke questions grounded in observations

Meme slide questions

Spontaneous humor capture

Promotion prompt

Blind spot questions

Demo reaction questions

Recency adaptation — recent

Recency adaptation — old

Structured output format

Both checker scripts referenced

Data attribution FAIL surfaced

Closing FAIL surfaced

Cut-lines FAIL surfaced

Rushed-closing recurring issue noted

Vague image prompt flagged (slide 5)

Schema validation confirmed up front

Big-idea singleton confirmed

Thesis preview/payoff ordering confirmed

Illustration coverage commentary

Report distinguishes FAIL vs FLAG vs INFO

outline.yaml validates against schema

style_anchor block present

Format-specific anchors

Visual continuity conventions documented

format field on every slide

EXCEPTION slides carry format_justification

image_prompt uses [STYLE ANCHOR] token

visual field present alongside image_prompt

Prompt richer than visual description

Visual continuity: FIG numbering

Visual continuity: progressive form

Standard outline structure preserved

Web-search precedes generation

No spurious model-change proposal

Proceeds to generation without re-litigating

Generation uses the outline's declared model

Web-search precedes generation

Outline model recognized as out-of-current-list

Explicit decision point before generation

No silent generation with stale model

Request A routes to Strategy mode

Request B routes to Generation mode

Request C routes to Thumbnail mode

Step references resolve against the skill's canonical sequence

Request B includes builds because outline has Builds blocks

Request B includes apply-to-deck because .pptx exists

Request A finishes after Step 2

No mode confusion across requests

4-tier structure present

Signature tier from profile

Contextual tier from profile

Shake It Up has 1-2 items

[RECURRING] tag used

[CONTEXTUAL] tag used

New to You from never_used

Actionable recommendations

Skips static files

Skips conflict copies

Skips template files

Background colors extracted

Per-slide breakdown present

Global design statistics

Shape type detail

Font extraction

Footer detection

Recursive directory scan

Valid JSON output

Build specification format

Build step 0 is empty frame

Final build matches the full slide

Backwards-chaining workflow described

Build naming convention

Build slides count toward budget

Speaker notes on final step only

BLANK layout for all builds

Explicit preservation instructions

Safety suffixes mentioned

Parent slide replaced by build sequence

Talk slug read from outline.yaml

Shownotes URL constructed from pattern + outline.yaml slug

Bitly custom back-half is the talk slug

QR encodes bitly URL, not raw shownotes URL

generate-qr.py used, not qrcode library directly

Command uses --shownotes-url (not --short-url)

Tracking DB update mentioned

No silent degradation to raw URL

Agent surfaces missing shortener config

Agent asks user to choose a shortener

Agent does NOT generate QR with raw URL without asking

Shownotes URL uses slug from outline.yaml (talk.slug)

Talk slug matches spec exactly

Agent references generate-qr.py as the tool

Agent distinguishes missing config from opt-out

Top-level DB structure

Config captures source directories

Talk fields extracted

URL fields parsed

New talks default to pending

No-sources talk flagged as unprocessable

Slide source determination

Static files skipped

Conflict copies skipped

Template files skipped

PPTX-to-talk matching

Scan report with counts

Recommended aesthetic is comic_book

Decision cites the precedence step that produced the answer

Decision quotes the matched field value

No side-by-side comparison forced

Speaker photo path from profile

style_preference and title_position applied

Brand colors applied

Step 1 not silently skipped

Title argument is the speaker-confirmed hook verbatim

Iterate not restart strategy

Style changed to split_panel

Title position changed to top

Title text unchanged

Slide image unchanged

Face issue addressed via style variant, not preservation language

generate-thumbnail.py used

Slide image from illustrations directory

Original vs revised comparison

Speaker photo path unchanged

Wide-angle detection

Hash threshold recommendation

Slide region crop suggestion

Multi-language transcript handling

Whisper hallucination detection

Transcript source tracking

Speaker identity verification

Recording type classification

Structured diagnostics output

Clean recording passes cleanly