Insert slide memes

Insert Slide Memes

Use humor as pacing, analogy, reaction, or callback. Keep the deck's argument primary.

Resolve references/ and scripts/ relative to the directory containing this SKILL.md. Do not assume the current working directory is the plugin root.

Select a mode

Honor an explicitly requested mode. Otherwise:

• Choose postprocess when the user supplies a completed HTML deck or asks to revise an existing deck.
• Choose plan-and-build when the user supplies a topic, brief, source material, or outline and asks to create HTML slides with memes.
• If both an outline and HTML exist, choose the mode matching the requested stage. Do not require finished HTML for plan-and-build.
• State the selected mode in the first work update.

Select a rights mode

Rights mode is separate from postprocess or plan-and-build. Honor an explicit choice and record it as root-level rights_mode in the JSON plan.

• strict is the default. Use it for any client, paid, shared-file, exported, recorded, or public use, and whenever the delivery context is unknown. A selected asset needs a license, permission, public-domain basis, or documented statutory exception.
• practical is available only for a one-off live-internal presentation with distribution: internal, no file sharing, no client or paid audience, and no export, recording, or publication. It permits a selected asset after the compact practical review in the rights contract; it does not declare the use lawful.
• Switch the whole plan to strict before any broader use. Do not silently carry a practical approval into a new context.

Shared selection rules

1. Establish audience, setting, language, formality, shared references, humor tolerance, and every intended use mode. Treat an unverified shared reference as absent; do not infer fandom, gaming, political, or subculture affinity from profession, nationality, age, or technical fluency.
2. Read references/meme-playbook.md.
3. Read references/rights-clearance.md. Do not download, copy, embed, or insert an asset until it has either the strict legal basis or the practical review required by the selected rights mode.
4. Read references/user-provided-memes.md when the user supplies any asset, URL, template name, caption, or placement instruction.
5. Define the communicative job before searching for an image.
6. For every unresolved placement, start globally established meme discovery with Imgflip's template and active-caption ecosystem. Use it to compare comic execution, variation, and fatigue; use Know Your Meme or another maintained contextual source afterward to verify established meaning and origin. When the target culture is poorly represented on Imgflip, use a regional source first and record why. Explore globally common, regional or language-specific, and relevant professional or community meme ecosystems as coverage, not a quota.
7. Apply every hard gate in the playbook before scoring or sourcing an asset. A caption cannot turn promotional art, wallpaper, stock imagery, fandom art, or an unrelated illustration into a meme.
8. Prefer an established, recognizable template whose familiar meaning performs part of the joke. Write a new slide-specific caption for that template. Do not generate a new visual meme format or AI-original illustration unless the user explicitly asks.
9. Record the rights mode, discovery route and page, humor evidence, complete per-dimension score, recognition basis, identity signal, source roles, legal basis or practical review, use modes, additional-rights checks, attribution location, and gate results in a machine-readable JSON plan. Run python3 <skill-root>/scripts/audit_meme_plan.py path/to/meme-plan.json --strict before downloading or inserting any asset.
10. Localize the caption while preserving the template's familiar meaning.
11. Keep active-humor discovery, semantic history, original provenance, asset location, legal evidence, and attribution as separate facts. An Imgflip discovery page is not semantic or legal evidence, and a Know Your Meme history page is not the asset source. For Korean candidates, follow the Korean source hierarchy and static-image workflow in the playbook. Download only a candidate that already passed selection; never hotlink third-party images or capture frames from video.
12. Use the adaptive density rules in the playbook instead of a fixed deck-wide count. Treat the audience-specific range as a soft ceiling, not a quota; zero memes is valid, and long decks may exceed three when enough high-value placements pass every gate. Normally leave at least five non-meme content slides between distinct meme beats, account for section dividers and live demonstrations that already reset attention, and use at most one meme per slide. Honor an explicit user count while warning if it materially weakens the deck.
13. Reject a searched candidate that decorates, repeats the slide, needs explanation, depends on an unverified subculture, implies an unsupported presenter identity, or weakens the presenter's credibility. Keep any asset with unresolved rights provisional, including a user-supplied asset, until it passes the review required by the selected rights mode.

Mode: postprocess

1. Inspect the complete deck, assets, framework, slide boundaries, navigation, stable identifiers, and intended viewport.
2. Identify high-value release, analogy, callback, and transition moments without changing the deck's argument.
3. Create the machine-readable meme plan using the fields in Meme plan.
4. Use reviewed user-supplied assets and instructions first. Search and hard-gate candidates only for unresolved placements. Keep unverified user assets provisional until they pass the selected rights mode. Audit the plan before downloading any selected asset.
5. Insert semantic markup and scoped CSS. Preserve the existing slide system, stable IDs, visual language, and keyboard behavior.
6. Choose the packaging method:
   • Default to relative local assets beside the deck.
   • Use Base64 data: URLs when the user requests a single self-contained HTML file.
   • Preserve data-meme-source metadata in either case.
7. Render every changed slide and its neighbors at the target viewport and one smaller viewport.
8. Verify keyboard navigation, image loading, overflow, and browser console output.
9. Run python3 <skill-root>/scripts/audit_memes.py path/to/deck.html --plan path/to/meme-plan.json --strict.
10. Report changed slides, why each joke belongs, asset sources, reuse status, and packaging.

Mode: plan-and-build

Read references/plan-and-build.md, then:

1. Establish the presentation goal, audience, setting, duration, tone, source material, and delivery format.
2. Build a content-first outline. Give every slide a narrative job and one primary message.
3. Mark only the moments that benefit from release, analogy, callback, or transition. Do not place memes at fixed intervals.
4. Add provisional meme briefs to the outline before choosing exact assets. Keep content slides understandable if every meme is removed.
5. Apply user-supplied assets and instructions to the relevant outline records. Search and hard-gate recognizable candidates only for unresolved placements after the surrounding argument is clear. Record the complete gate, score, audience, identity, source, layout, and risk fields.
6. Treat outline approval already required by the presentation workflow as approval of the meme beats, not as approval of asset rights. Resolve every material tone, reputational, or rights risk separately.
7. Generate the complete HTML deck with the selected meme moments included. Use stable slide identifiers from the first build.
8. Audit the machine-readable plan before asset download, then apply the same packaging, markup, guardrail, rendering, and HTML audit requirements as postprocess.
9. Replace or drop any meme that fails the two-second recognition test in the rendered deck.
10. Report planned versus final meme placements and any candidate changed during visual QA.

Meme plan

For each proposed placement record:

• unique plan ID and slide identifier
• slide identifier and narrative job
• role: reaction, analogy, callback, or transition
• earlier setup slide ID when the role is callback
• intended audience response
• origin: user-provided or searched
• user-locked fields such as asset, template, caption, or placement
• template name and asset kind
• recognition basis and concrete audience evidence; never assumed for a selected searched candidate
• identity signal level, domain, and explicit user approval when material
• every hard-gate result from the playbook
• all seven score dimensions, total, ubiquity penalty, and adjusted score
• discovery route: imgflip-first, regional-first, or fallback-other
• direct discovery page, concrete humor evidence, and a fallback reason when the route is not imgflip-first
• candidates considered across cultures and why the winner is most recognizable
• caption or setup, preferably under 12 words
• separate semantic, original, asset, legal-evidence, and HTML attribution sources
• plan-level rights mode: strict or practical
• rights status, intended use modes, distribution scope, and attribution location
• strict legal basis, jurisdiction, evidence date and scope; or the complete practical review
• moral-rights, portrait/publicity, and trademark checks
• attribution text, URL, and location
• layout and primary-content tradeoff
• risk: audience fit, ambiguity, stereotype, copyright, or dated reference
• status: provisional, selected, or dropped

Use the JSON field names and example in references/plan-and-build.md. A Markdown summary may accompany the JSON, but it does not replace the audited plan.

HTML contract

Use this structure when the deck permits it:

<figure
  class="slide-meme"
  data-meme-plan-id="m01"
  data-meme-role="reaction"
  data-meme-template="RECOGNIZED_TEMPLATE_NAME"
  data-meme-source="SOURCE_URL_HERE"
  data-meme-origin="searched"
>
  <img
    src="assets/memes/recognized-template.jpg"
    alt="Description of the recognizable reaction shown in the meme"
  />
  <figcaption>Localized caption that completes the joke</figcaption>
  <a
    class="meme-attribution"
    data-meme-attribution-location="on-slide"
    href="ATTRIBUTION_URL_HERE"
  >
    Creator — Work, license
  </a>
</figure>

Map data-meme-plan-id, data-meme-role, data-meme-template, data-meme-source, and data-meme-origin from the audited plan. Set data-meme-source and the .meme-attribution link to the plan's attribution_url; never substitute a semantic page, aggregator, or user-provided marker. Use data-meme-plan-id on an attribution placed outside the figure and map data-meme-attribution-location from the plan. Put credits in a slide classed as credits, credits-slide, references, or sources; put speaker-note attribution in a speaker-notes or data-speaker-notes container. Keep meaningful alternative text distinct from the joke caption. Use alt="" only for a truly decorative image.

Scope styles under the deck or .slide-meme namespace:

.slide-meme {
  margin: 0;
  max-width: min(34vw, 30rem);
}

.slide-meme img {
  display: block;
  width: 100%;
  max-height: 42vh;
  object-fit: contain;
}

.slide-meme figcaption {
  margin-top: 0.4rem;
  font-size: clamp(0.85rem, 1.4vw, 1.1rem);
  line-height: 1.2;
}

Adapt values to the deck. Do not introduce global img, figure, or typography rules.

Guardrails

• Do not put a meme on the title, agenda, legal/compliance material, sensitive personal stories, layoffs, safety incidents, or solemn conclusions unless explicitly requested and clearly appropriate.
• Do not target protected traits, appearance, disability, victims, junior staff, customers, or identifiable coworkers.
• Avoid political, sexual, violent, insulting, or profanity-heavy material by default.
• Treat gaming, sports-team, entertainment-franchise, political, religious, and other fandom imagery as a material presenter-identity signal. Do not use it for a searched candidate without concrete audience evidence and explicit user approval.
• Reject promotional art, wallpapers, press images, stock images, and unrelated illustrations as searched memes even when their slogan or subject creates a topical word association.
• Do not treat popularity, search availability, user provision, source attribution, or an internal audience as permission.
• Use public Imgflip examples to understand format rhythm and current variation; do not copy another user's distinctive caption verbatim as the deck's punchline.
• Do not select or copy an asset with rights_status: unclear or user-provided-unverified. In strict mode, use a cleared alternative or document a statutory-exception analysis. In practical mode, complete the practical review and change the status to practical-reviewed.
• Recheck scope when the deck gains a client audience, fee, file share, export, recording, or online publication; each of these requires strict mode.
• Never invent an image source, license, quotation, or attribution.
• Do not let a meme shrink body text, cover controls, cause overflow, or become necessary to understand the slide.
• Preserve a usable deck if images fail to load.

Verification

Check:

• no cropping of faces, captions, or key visual details
• no overlap at the target viewport and one smaller viewport
• sufficient caption contrast and readable type
• no remote hotlinks; local or valid embedded assets only
• meaningful alt text, source metadata, and attribution at the planned location
• exact correspondence with a selected record in the audited meme plan
• restrained density and no more than one meme per slide
• coherent visual treatment
• no autoplay, flashing animation, or layout shift

The audit script performs structural checks, not visual judgment. Rendering and human-context review remain mandatory.