Presentation Skill

Presentation Skill

Repo-native PowerPoint skill for editable, aligned, QA-checked .pptx decks. The model plans the story and visual strategy; scripts handle fragile rendering, staging, and verification.

When to invoke this skill

Trigger this skill on requests to:

• build a one-off PowerPoint .pptx, slide deck, or presentation from a

single prompt (quick-deck mode, no workspace needed)

• generate a .pptx from a structured outline.json
• redesign, rebuild, or extend an existing deck
• create a lab, clinical, scientific, board, investor, or editorial deck
• run layout, design, or readability QA on a generated deck
• author or update a reusable deck workspace
• convert CSV/TSV/XLSX/JSON data into slide-ready figures and tables

Quick-deck mode is the right path for one-shot 5-10 slide decks. Workspace mode is for decks that will be iterated, audited, or rebuilt later. See ### Quick Deck and ### Saved Workspace under Workflow below.

Do not invoke it for text-only brainstorming where no deck artifact is needed, or for direct edits to a generated .pptx when its saved workspace source files are available (fix the source instead).

Skill name: presentation-skill. Compatibility aliases for fuzzy skill matching: powerpoint-deck-builder, pptx-skill, PowerPoint skill, PPTX skill, slide-deck generator, slides generator, deck builder, presentation generator, presentation maker, PowerPoint generator, Codex presentation skill, ChatGPT presentation skill, OpenAI agent presentation skill.

Non-Negotiables

• Do not write ad hoc inline python-pptx or pptxgenjs deck code. Author

outline.json and run repo scripts.

• Do not reinstall dependencies during deck generation. Missing dependency:

report it and stop.

• Do not skip QA for a deliverable deck. Use render-free QA when LibreOffice is

unavailable.

• For reusable/report decks, treat validate_planning.py warnings about

missing artifact registries, stale generated artifacts, figure rebuild commands, readability contracts, speed contracts, QA contracts, acceptance evidence, or execution phases as source-planning feedback to fix before final delivery.

• Fix source files (outline.json, plans, assets, renderer code), not mutated

.pptx artifacts.

• For a new topic, scaffold fresh. Do not clone another deck's structure as a

house style.

First Files To Read

• DESIGN.md: compact design contract, colors, hierarchy, alignment rules,

generated-image disclosure.

• references/outline_schema.md: accepted outline.json fields and variants.
• references/planning_schema.md: design_brief.json,

content_plan.json, and evidence_plan.json shape.

• references/deck_workspace_mode.md: saved-workspace workflow.
• references/agent_picker.md: concise picker guidance for agents and people.
• references/reference_script_patterns.md: reusable PPTX/DOCX/figure

patterns from stronger one-off scripts; read when improving routing, scientific decks, or companion document workflows.

• references/style_reference_catalog.md: publish-safe style-reference

policy for using public decks, screenshots, PPTX corpora, and synthetic recreations without bundling proprietary material.

• references/dynamic_design_and_subagents.md: bounded dynamic design

modulation plus when to use content, data, routing, critique, and QA subagents.

• references/visual_qa_prompt.md: visual-review packet and fresh-eyes inspection prompt.
• scripts/emit_deck_start_packet.py: preferred first-turn packet for

reusable decks; bundles a compact user question payload, reproducible design contract prompt, and staged subagent handoff commands.

• scripts/emit_deck_intake_prompt.py: optional user-facing intake prompt

for audience, style, palette, density, background, assets, sources, and constraints when personalization is underspecified.

• scripts/apply_deck_intake_answers.py: deterministic bridge from Codex

question-card answers to design_brief.user_intake, style_seed, notes, and planning-file source policy/asset posture.

• scripts/run_deck_start_intake_smoke.py: focused fast smoke for the

first-turn deck-start packet, versioned agent kickoff brief, durable intake answers, pre-answer readiness record_deck_intake_answers trigger, deterministic application, idempotence, and readiness deck_intake summary.

• scripts/emit_design_contract_prompt.py: reproducible style/structure

contract prompt to run immediately after the user request or intake answers, before authoring outline.json.

• scripts/apply_design_contract.py: deterministic bridge from a returned

deck_design_contract_v1 JSON packet to design_brief.json, content_plan.json, evidence_plan.json, asset_plan.json, and notes.md.

• scripts/run_design_contract_apply_smoke.py: focused fast smoke for the

design-contract prompt, intake-seeded choice-resolution enrichment, deterministic contract application, idempotence, planning validation, and readiness design_contract summary.

• scripts/emit_outline_authoring_prompt.py: contract-aware handoff prompt

for replacing starter outline.json with authored slides plus aligned content/evidence/asset plan edits after design_contract.json is applied.

• scripts/apply_outline_authoring_handoff.py: deterministic bridge from a

returned outline_authoring_handoff_v1 JSON packet to outline.json, content_plan.json, evidence_plan.json, asset_plan.json, and notes.md.

• scripts/run_outline_authoring_handoff_smoke.py: focused fast smoke for

the contract-aware outline prompt, deterministic handoff apply, source-plan alignment, readiness summary, PPTX build, and outline-aware render-free QA.

• scripts/run_reproducible_workflow_smoke.py: fast end-to-end smoke for the

deck-start, intake, design-contract, outline-handoff, strict render-free QA, and delivery-readiness handoff path.

• scripts/run_focused_workflow_checks.py: aggregate fast smoke runner for

the reproducible workflow, style-mix, layout polish, source-footer, and generated-data lanes; can run isolated checks concurrently while writing one compact JSON report with per-check durations and failure tails.

• scripts/extract_pptx_style.py: source-only PPTX/corpus style extractor

that emits observed header/footer/color/readability signals plus a reusable design_brief.json fragment for template-inspired decks.

• scripts/apply_pptx_style_fragment.py: deterministic bridge from extracted

PPTX style fragments to design_brief.json, renderer_treatments, style_import, and notes.md.

• scripts/style_reference_catalog.py: synthetic publish-safe style DNA and

content-treatment references for each preset, including title, comparison, chart, table, figure, dashboard, decision, and references behavior plus the style_reference_mix_plan_v1 prompt matcher and style_reference_layout_playbook_v1 variant map and style_reference_structural_motif_library_v1 motif grammar plus style_reference_metric_profile_v1 density/whitespace/object-bias metrics plus style_reference_content_recipe_library_v1 content-slot recipes used by design-contract and outline-authoring prompts.

• references/style_reference_sources.json and

scripts/style_reference_sources.py: publish-safe source-intake manifest and validator for public decks, screenshots, PPTX corpora, and template-gallery inspiration. Use this before adding or citing external style references; it records license notes, source-verification evidence, allowed extraction modes, forbidden materials, generic style observations, reusable slide pattern extracts, attribution posture, and per-preset synthetic reconstruction routes.

• references/style_inspiration_corpus.json and

scripts/style_inspiration_corpus.py: descriptor-only inspiration corpus for public design systems, presentation tooling, template indexes, and slide design heuristics. Use it as a scalable source-index layer for prompt-to-style routing and subagent scouting; it stores rights posture, extraction limits, layout/palette/typography descriptors, per-preset routes, and the safety rule that no raw decks, screenshots, logos, proprietary text, or distinctive copied geometry are bundled.

• references/large_style_corpus_sources.json,

references/large_style_corpus_catalog.json, and scripts/large_style_corpus.py: descriptor-only large public deck corpus layer for real-world style discovery. Use it to index 2,000+ public/open deck-like records, balance overlap, prioritize AI-agent/tooling examples, emit compact LLM context, and replace redundant-looking preset treatments with original synthetic structure. It stores only public URLs, repository/path metadata, inferred style families, and descriptors; never raw decks, screenshots, copied slide text, logos, or distinctive source geometry.

• references/style_token_atlas.json, references/style_grammar_index.json,

scripts/style_atom_router.py, scripts/apply_atom_composition.py, and scripts/workflow_atom_context.py: dynamic corpus-atom grammar layer. Use it when a deck needs the corpus to influence structure, treatment pools, density, and narrative arc while still leaving the current/future LLM free to make topic-fit choices. Normal deck-start, design-contract, and style/content router prompts now receive a normal_workflow_atom_context_v1 seed that can be accepted, refined through strict JSON, or skipped with a recorded reason. Apply only supported renderer fields.

• scripts/run_large_style_corpus_smoke.py: focused fast smoke proving the

large-corpus source manifest, descriptor schema, family coverage, compact context, and digest writer stay usable without network access.

• scripts/run_style_atom_router_smoke.py: focused fast smoke proving the

atom atlas, grammar templates, deterministic fallback, topic-aware ranking, strict JSON atom prompt, and applier replay ledger stay usable.

• scripts/build_large_style_corpus_contact_sheets.py: publish-safe visual

contact sheets generated from large-corpus descriptors. It draws synthetic family and source-record cards from catalog metadata only; it does not fetch, render, or screenshot source decks.

• scripts/run_large_style_corpus_leverage_smoke.py: focused smoke proving

the 2,000-record corpus is actually used by router prompts and contact-sheet evidence while preserving curated style-reference prompt tokens.

• scripts/design_catalog_selector.py: reusable v0.5 design-catalog selection

layer that maps prompts to topic cases, corpus families, treatment DNA, generated data recipes, and reproducible release-evidence scope.

• scripts/build_random_topic_comparison_decks.py: release-evidence builder

for eight random synthetic topics, each built as a baseline deck and a descriptor-corpus-guided deck across eight visibly different corpus families, plus rendered contact sheets, a gallery deck, generated data/chart/table artifacts, manifest metrics, structural-delta signatures, and concise release notes.

• scripts/run_random_topic_comparison_smoke.py: reproducibility smoke that

rebuilds the random-topic comparison evidence in a temporary directory and fails if the 16 decks, gallery, manifest, release notes, nonblank contact sheets, corpus-context metrics, eight-family coverage, three generated data examples, structural QA gates, zero visual/readability warnings, or baseline-vs-corpus structural deltas are missing.

• scripts/build_style_reference_gallery.py: generated reference-gallery

decks that turn the synthetic style catalog into actual title, dashboard, comparison, chart, table, figure, decision, and references slides without bundling external template assets. The gallery summary records per-preset variant sequences, structural playbook signatures, structural motif signatures, content-recipe signatures, renderer-treatment signatures, treatment buckets, rendered contact-sheet paths, per-treatment contact-sheet paths, per-preset contact-sheet collections for overview, data_evidence, and decision_sources, and QA totals for release evidence.

• scripts/run_style_reference_gallery_smoke.py: focused fast smoke proving

all-preset style-reference gallery decks keep unique content signatures, required treatment coverage, publish-safe metadata, and clean render-free QA.

• scripts/run_style_reference_resolution_smoke.py: all-preset build-time

resolver smoke proving identical generic evidence slides resolve into preset-specific build/outline_resolved.json variants with unique treatment signatures.

• scripts/run_style_content_router_smoke.py: focused fast smoke proving the

style/content router prompt includes ranked style-reference matches, mix-plan context, layout playbooks, content recipes, renderer treatment pools, and the optional first-class atom composition route for prompt-to-reference routing.

• scripts/run_style_reference_release_evidence_smoke.py: optional rendered

release-evidence smoke proving the all-preset gallery has clean QA, unique content signatures, visual-diversity hashes, fingerprinted contact sheets, rendered slide evidence for every required treatment key, and per-treatment rendered thumbnail and coarse-layout signatures so repeated-looking treatment families are visible and blocked in release evidence.

• scripts/run_style_reference_starter_smoke.py: all-preset starter smoke

proving initialized style_reference_starter_outline_v1 workspaces keep distinct scaffold signatures, publish-safe source metadata, starter assets, and clean render-free QA.

• scripts/style_treatment_profiles.py: reusable preset treatment profiles

for supported heading/accent, footer, chart, and figure/table pools across all loadable presets. Profiles include a style_reference_catalog_v1 reference so agents get more than surface-level chrome variation.

• scripts/build_header_variant_gallery.py: visual fixture builder for

checking clean content-header variants across all presets.

• scripts/run_header_variant_gallery_smoke.py: focused fast smoke proving

all presets build actual header-variant gallery decks with clean render-free QA counts.

• scripts/run_rendered_header_gallery_smoke.py: optional slower rendered

smoke proving all presets render the header-variant gallery to nonblank slide images, write a contact sheet, and pass lab-report visual review.

• scripts/run_rendered_data_delivery_smoke.py: optional slower end-to-end

smoke proving a lab/data deck carries deck-start intake, a design contract, a rendered figure/chart/table artifact triplet, strict rendered QA, visual-review contact sheet, generated-slide density floors, speed timings, repeat-build source/artifact/quality stability, and final delivery-readiness ready status.

• scripts/run_style_mix_repro_smoke.py: focused fast smoke for deterministic

style-mix/header/footer treatment resolution.

• scripts/run_layout_polish_handoff_smoke.py: focused fast smoke proving

saved QA whitespace/readability warnings become exact readiness and advance_workspace.py source-edit handoffs.

• scripts/run_readability_contract_smoke.py: focused fast smoke proving

rendered chart/table readability-contract warnings preserve measurements, slide IDs, and exact source-edit handoffs.

• scripts/trim_image_whitespace.py: helper for generated plots with large

exterior whitespace before insertion into figure-first layouts.

• scripts/compact_source_footers.py: source-only helper that rewrites

over-budget source-line footer provenance into short IDs and a final editable References table slide, splitting reference rows across slides when needed.

• scripts/run_source_footer_compaction_smoke.py: focused fast smoke for

preflight footer-budget warnings, readiness/advance compaction, reference slide creation, and idemp

…

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: sirilsengolraj-source
• Source: sirilsengolraj-source/presentation-skill
• License: MIT

Install and usage instructions live in the source repository linked above.