presentation-skill v0.9.0

presentation-skill v0.9.0

v0.9 moves composition grammar from title treatment into the full deck.
Evidence, comparison, chart, table, decision, and references slides now resolve
through the same grammar contract as covers and section turns.

What changed

• Added renderer_role_contracts_v2: eight color-independent structural
  systems for each of eight slide roles.
• Added a declarative role-layout compiler with normalized semantic slots and
  bounded primary, alternate, and dense slide overrides.
• Kept text, charts, tables, figures, and diagrams editable in PowerPoint.
• Made new workspaces persist and apply v2 through deck start, design contract,
  style routing, outline authoring, and workspace build.
• Kept v1-only workspaces pinned until the explicit, idempotent migration
  command is run.
• Replaced readiness-report command execution with a typed action registry that
  validates operations, flags, and workspace-contained paths.

See it

• Narrative structures
• Evidence and data structures
• Decisions and sources
• Editable 64-slide gallery

The gallery uses one urban-heat topic across all eight grammar families so the
differences are visible without changing the subject.

Upgrade an existing workspace

python3 scripts/upgrade_renderer_role_contracts_v2.py \
  --workspace decks/my-deck

Running the command again is a no-op. Workspaces without the v2 contract keep
their v1 rendering behavior.

v0.8.0

v0.8.0

This release packages presentation-skill as a Codex plugin marketplace entry
so people can install it from the Codex plugin browser instead of manually
copying a skill folder.

What changed

• Added .agents/plugins/marketplace.json with a Presentation Skill
  marketplace entry.
• Added plugins/presentation-skill/.codex-plugin/plugin.json with public
  plugin metadata, screenshots, keywords, starter prompts, and MIT license
  metadata.
• Added a synced plugin skill snapshot under
  plugins/presentation-skill/skills/presentation-skill.
• Added scripts/sync_plugin_snapshot.py so future releases can refresh the
  plugin snapshot from the root skill deterministically.
• Updated README install docs with codex plugin marketplace add usage.

Install

codex plugin marketplace add siril9/presentation-skill --ref v0.8.0

Then open /plugins in Codex and install presentation-skill from the
Presentation Skill marketplace.

Validation

• python3 scripts/sync_plugin_snapshot.py
• python3 <plugin-creator>/scripts/validate_plugin.py plugins/presentation-skill
• python3 -m json.tool .agents/plugins/marketplace.json
• python3 -m json.tool agents/discovery.json
• npm run check:python
• npm run check:node

Share and curation

Copy-ready community posts and an OpenAI curation-request draft are in docs/PROMOTION.md.

v0.7.0

v0.7.0

This release makes atom composition a first-class optional handoff in the normal
workspace workflow, not only in release comparison builders.

What changed

• Added normal_workflow_atom_context_v1, generated by
  scripts/workflow_atom_context.py.
• init_deck_workspace.py now seeds atom context into starter workspace files.
• emit_deck_start_packet.py now includes atom_composition as an active,
  reproducible route with accept/refine/skip semantics.
• emit_design_contract_prompt.py and emit_style_content_router.py now expose
  the atom seed, preferred variants, narrative arc, strict JSON refinement
  prompt, and supported renderer deltas.
• apply_deck_intake_answers.py and apply_design_contract.py preserve atom
  choices in choice_resolution, style_system, and reproducibility replay
  metadata.
• README imagery now leads with renderer-variant and style-family proof boards,
  then keeps the clean Codex-native vs updated-skill comparison as supporting
  evidence.

Release Evidence

• Clean README hero:
  decks/native-vs-latest-random-topics-20260623/readme_images/presentation_skill_variant_proof.png
• Style-family proof board:
  decks/native-vs-latest-random-topics-20260623/readme_images/presentation_skill_style_family_proof.png
• Clean three-topic comparison:
  decks/native-vs-latest-random-topics-20260623/readme_images/codex_native_vs_updated_clean_three_topics.png

Validation

• npm run check:style-atoms
• npm run check:deck-start
• npm run check:design-contract
• npm run check:style-router

presentation-skill v0.6.0

presentation-skill v0.6.0

This release turns the 2,000-record descriptor-only public deck corpus into reproducible deck evidence, not only a catalog.

What changed

• Expanded the random-topic baseline-vs-corpus comparison builder to 8 synthetic topics and 16 case decks.
• Added a reusable design-catalog selection layer that records primary family, treatment tags, data recipe, and structure intent.
• Added manifest fields that prove corpus-guided outlines carry large-corpus context while baseline outlines do not.
• Added structural sequence signatures so corpus leverage can be checked as slide grammar, not only color/header chrome.
• Added generated data-artifact examples with CSV, editable chart JSON, editable table JSON, artifact manifest, and analysis summary files.
• Added compact rendered contact sheets and a gallery deck for quick visual inspection.
• Kept all evidence publish-safe: synthetic slide content plus descriptor-only corpus metadata; no external decks, screenshots, logos, copied text, or copied geometry.

Evidence

• Manifest: decks/random-topic-corpus-comparison-v0.6.0-20260629/manifest.json
• Gallery deck: decks/random-topic-corpus-comparison-v0.6.0-20260629/comparison-gallery/build/random-topic-corpus-comparison-gallery.pptx
• Overview contact sheet: decks/random-topic-corpus-comparison-v0.6.0-20260629/contact_sheets/all_topics_baseline_vs_corpus.png
• Pair contact sheets: contact_sheets/*_baseline_vs_corpus.png
• Builder: scripts/build_random_topic_comparison_decks.py
• Smoke gate: python3 scripts/run_random_topic_comparison_smoke.py

Validation snapshot

• Automated smoke gate: pass
• Decks generated: 16
• Topics generated: 8
• Unique corpus families: 8 / 8
• Generated data examples: 4 / 3
• Corpus-guided cases: 8
• Outlines with corpus context: 8
• Warning budget: 0 visual, 0 readability, zero layout blockers
• Pair structural deltas: 8 / 8
• Nonblank contact sheets: 9 / 9

Residual risk

This is corpus-leverage release evidence. It is not a full published-GitHub-baseline audit; that broader comparison remains covered by earlier release evidence galleries.

presentation-skill v0.5.0

presentation-skill v0.5.0

This release turns the 2,000-record descriptor-only public deck corpus into reproducible deck evidence, not only a catalog.

What changed

• Expanded the random-topic baseline-vs-corpus comparison builder to 8 synthetic topics and 16 case decks.
• Added a reusable design-catalog selection layer that records primary family, treatment tags, data recipe, and structure intent.
• Added manifest fields that prove corpus-guided outlines carry large-corpus context while baseline outlines do not.
• Added structural sequence signatures so corpus leverage can be checked as slide grammar, not only color/header chrome.
• Added generated data-artifact examples with CSV, editable chart JSON, editable table JSON, artifact manifest, and analysis summary files.
• Added compact rendered contact sheets and a gallery deck for quick visual inspection.
• Kept all evidence publish-safe: synthetic slide content plus descriptor-only corpus metadata; no external decks, screenshots, logos, copied text, or copied geometry.

Evidence

• Manifest: decks/random-topic-corpus-comparison-v0.5.0-20260622/manifest.json
• Gallery deck: decks/random-topic-corpus-comparison-v0.5.0-20260622/comparison-gallery/build/random-topic-corpus-comparison-gallery.pptx
• Overview contact sheet: decks/random-topic-corpus-comparison-v0.5.0-20260622/contact_sheets/all_topics_baseline_vs_corpus.png
• Pair contact sheets: contact_sheets/*_baseline_vs_corpus.png
• Builder: scripts/build_random_topic_comparison_decks.py
• Smoke gate: python3 scripts/run_random_topic_comparison_smoke.py

Validation snapshot

• Automated smoke gate: pass
• Decks generated: 16
• Topics generated: 8
• Unique corpus families: 8 / 8
• Generated data examples: 4 / 3
• Corpus-guided cases: 8
• Outlines with corpus context: 8
• Warning budget: 0 visual, 0 readability, zero layout blockers
• Pair structural deltas: 8 / 8
• Nonblank contact sheets: 9 / 9

Residual risk

This is corpus-leverage release evidence. It is not a full published-GitHub-baseline audit; that broader comparison remains covered by earlier release evidence galleries.

v0.3.0 - Large public deck style corpus

v0.3.0 - Large public deck style corpus

Adds a descriptor-only public deck corpus for real-world style discovery and LLM routing.

Added

• 2,000 public/open deck-like records in references/large_style_corpus_catalog.json.
• Query/source manifest in references/large_style_corpus_sources.json covering GitHub code search and repository-tree indexing routes.
• Human-readable corpus digest in references/large_style_corpus_catalog.md.
• scripts/large_style_corpus.py for discovery, resume, validation, overlap balancing, family-floor rebalancing, compact LLM context, and digest generation.
• scripts/run_large_style_corpus_smoke.py plus npm run check:large-style-corpus.
• Style/content router integration so scouts can use compact large-corpus context after the curated synthetic style-reference match plan.

Corpus shape

• 2,000 descriptor-only records.
• 740 unique repositories.
• 13 style families with at least 10 records each.
• 135 AI/agent/tooling signal records.
• No raw third-party PPTX/PDF files, screenshots, copied slide text, logos, or distinctive source geometry are bundled.

Validation

• python3 scripts/large_style_corpus.py --validate --min-records 2000 --min-family-records 10
• npm run check:large-style-corpus
• npm run check:style-router
• npm run check:python

presentation-skill v0.2.0

v0.2.0 Release Notes

Summary

This release adds a descriptor-first style-reference layer for richer,
reproducible PowerPoint generation. The skill now gives agents a browseable
design-memory path instead of relying only on preset colors, headings, and
borders.

What Changed

• Added a descriptor-only inspiration corpus for public design systems,
  presentation tooling, template indexes, and slide-design heuristics.
• Added corpus validation with rights posture, allowed extraction modes,
  forbidden materials, preset routes, and subagent safety rules.
• Extended the style/content router prompt so a scout can use corpus routes,
  selected source descriptors, and preset contact-sheet use cases before
  locking a design contract.
• Added per-preset contact-sheet collections for all 13 presets:
  overview, data_evidence, and decision_sources.
• Extended release evidence so rendered gallery checks fail if a preset is
  missing its contact-sheet collection or expected treatment thumbnails.
• Added compact release evidence for browseable style comparison without
  bundling external decks, screenshots, logos, proprietary text, or copied
  slide geometry.

Evidence

• Presets covered: 13
• Per-preset contact sheets: 39
• Global/treatment contact sheets: 11
• Required per-preset use cases: overview, data_evidence,
  decision_sources
• Release evidence gate: passed
• Render-free QA totals: 0 overflow, 0 overlap, 0 design warnings,
  0 visual warnings, 0 placeholder hits

Validation

• python3 scripts/style_inspiration_corpus.py --validate
• npm run check:style-corpus
• npm run check:style-router
• npm run check:style-reference-gallery
• npm run check:style-reference-release
• npm run check:python
• npm run check:focused -- --jobs 4
• git diff --check

presentation-skill v0.1.6

presentation-skill v0.1.6 Release Notes

This is the public v0.1 release-line package for the updated v1.1 showcase
workflow.

This release focuses on structured, reproducible, taste-constrained slide generation rather than one-shot slide rendering.

Why this release is worth showing

• Source-first deck workspaces: outline.json, design_brief.json, content_plan.json, evidence_plan.json, and asset_plan.json travel with the PPTX.
• Reproducible style decisions: stable style seeds, supported treatment pools, and resolved heading/footer variants.
• Cleaner lab/report slides: tuned 15 pt comparison body copy, compact source-line footers, bottom-right page numbers, readable tables, and evidence-first layouts.
• QA-led delivery: rendered slides are checked for overflow, overlap, geometry, placeholder text, visual warnings, and design warnings.
• Data/artifact path: local CSV/Excel/JSON inputs can become reusable figures, chart specs, and summary tables.

Release evidence

• Gallery deck: decks/release-v1.1-showcase-20260619/comparison-gallery/build/presentation-skill-v1-1-release-gallery.pptx
• Contact-sheet PNGs: decks/release-v1.1-showcase-20260619/comparison-gallery/assets/comparisons
• Comparison matrix: native bundled skill vs published GitHub v1 vs local v1.1.
• 13 style cases: Lab report, Risk memo, Startup, Editorial, Policy, Ops dashboard, Clinical exec, Arctic minimal, Midnight neon, Investor, Lavender ops, Terracotta, Editorial minimal.
• The build produced 39 comparison decks plus one gallery deck; the repo keeps
  the gallery deck, PNG contact sheets, manifest, and builder script as the
  compact release evidence.
• The comparison data is synthetic and is intended only to show generation
  behavior across styles.

Verification

• npm run check:focused passed for the release commit and the 15 pt body-copy tune.
• scripts/build_release_showcase.py generated the 13-style comparison matrix
  and gallery deck.
• Detailed per-deck counts remain available in
  release_showcase_manifest.json for audit/debugging without crowding the
  release notes.