clayz-presentation-skills

KO | EN

gitlite — search

#docker #gpt #rust #windows #cli #framework #agent #llm #mysql #svg

clayz-presentation-skills

(★ 9)

model-agnostic presentation production skills by Clayz

File Explorer

ZIP Download

- plugin.json
- - bug_report.yml
  - compatibility.yml
  - config.yml
  - feature_request.yml
  - output_quality.yml
- - ci.yml
  - pages.yml
  - release.yml
- - bcg-demo.png
  - clayz-overview.png
  - mckinsey-demo.png
- clayz-mark.svg
- - asymmetric-evidence-to-decision.json
  - flow-with-governed-checkpoint.json
  - hero-with-evidence-cluster.json
  - README.md
- - equal-weight-flattening.json
  - localization-overflow.json
  - README.md
  - semantic-visual-mismatch.json
  - sequence-order-loss.json
- - evidence-with-implication.json
  - README.md
  - sequence-with-checkpoint.json
- - README.md
  - synthetic-evidence-comparison.json
  - synthetic-evidence-decision.json
  - synthetic-evidence-opening.json
- - README.md
  - synthetic-evidence-to-decision.json
- provider-manifest.json
- README.md
- records.jsonl
- component-versions.json
- config.schema.json
- default.json
- architecture.md
- chatgpt-personal-light.md
- chatgpt-personal-light.zh-CN.md
- configuration.md
- feedback-benchmark-release-readiness.md
- feedback-benchmark-release-readiness.zh-CN.md
- index-runtime.md
- index-runtime.zh-CN.md
- knowledge-runtime.md
- knowledge-runtime.zh-CN.md
- layout-contracts.md
- layout-contracts.zh-CN.md
- pattern-dataset-library.md
- pattern-dataset-library.zh-CN.md
- personal-extension.md
- personal-extension.zh-CN.md
- provenance-policy.md
- release-packages.md
- release-packages.zh-CN.md
- releasing.md
- runtime-architecture.md
- runtime-architecture.zh-CN.md
- source-adoption.md
- source-adoption.zh-CN.md
- - brief.json
  - layout-tree.json
  - README.md
  - render-manifest.json
  - source-data.csv
- - feedback-index-report.json
  - learning-admissions.jsonl
  - learning-records.jsonl
  - legacy-admissions.jsonl
  - legacy-asset-registry.jsonl
  - legacy-migration-report.json
  - README.md
  - retrieval-benchmark-report.json
  - retrieval-benchmark.json
- - builtin-catalog.jsonl
  - external-metadata.jsonl
  - local-library.jsonl
  - public-receipt.json
  - public-request.json
  - README.md
- - comparison-compilation.json
  - comparison-instance.json
  - comparison-receipt.json
  - comparison-request.json
  - comparison-resolution.json
  - comparison-resolved.json
  - README.md
  - unresolved-receipt.json
  - unresolved-request.json
  - unresolved-resolution.json
- - comparison-composition-plan.json
  - comparison-failure-receipt.json
  - comparison-pattern-receipt.json
  - comparison-request.json
  - comparison-resolution.json
  - metadata-dataset-export.json
  - README.md
  - unresolved-receipt.json
  - unresolved-request.json
  - unresolved-resolution.json
- .nojekyll
- app.js
- case-manifest.json
- index.html
- README.md
- styles.css
- - .gitkeep
- - admitted-references.jsonl
  - asset-registry.jsonl
  - schema.md
- - NAVIGATION.md
- README.md
- - composite-skill.md
- - artifact-envelope.schema.json
  - capability-resolution.schema.json
  - component-version-report.schema.json
  - composition-pattern-request.schema.json
  - composition-pattern-resolution.schema.json
  - composition-pattern.schema.json
  - composition-plan.schema.json
  - failure-pattern.schema.json
  - feedback-index-report.schema.json
  - handoff.md
  - index-execution-evidence.schema.json
  - index-record.schema.json
  - knowledge-learning.md
  - layout-compilation.schema.json
  - layout-contract-instance.schema.json
  - layout-contract-request.schema.json
  - layout-contract-resolution.schema.json
  - layout-contract.schema.json
  - layout-tree.schema.json
  - learning-admission.schema.json
  - learning-record.schema.json
  - legacy-index-migration-report.schema.json
  - metadata-dataset-export.schema.json
  - personal-extension-profile.schema.json
  - personal-extension-runtime.schema.json
  - provider-manifest.schema.json
  - reference-record.schema.json
  - release-readiness.schema.json
  - render-manifest.schema.json
  - resource-inventory.schema.json
  - retrieval-benchmark-report.schema.json
  - retrieval-benchmark.schema.json
  - retrieval-receipt.schema.json
  - retrieval-request.schema.json
  - runtime-preflight.schema.json
  - sequence-record.schema.json
  - supervised-delivery-manifest.schema.json
  - version-private-learning-audit.schema.json
- - __init__.py
  - benchmark.py
  - learning.py
  - migration.py
  - readiness.py
- - __init__.py
  - capability.py
  - constants.py
  - io.py
  - provider.py
  - retrieval.py
  - runtime.py
  - utils.py
  - validation.py
- - __init__.py
  - compile_layout_contract.py
  - layout_contract.py
  - solve_relative_layout.py
  - validate_layout_contract.py
- - __init__.py
  - compile_composition_pattern.py
  - export_metadata_dataset.py
  - pattern_library.py
  - validate_pattern_library.py
- - __init__.py
  - resolver.py
- - __init__.py
  - preflight.py
- - audit_ppt_legibility.py
  - audit_pptx_size.py
  - audit_visual_rhythm.py
  - compare_package_to_pptx.py
  - config_policy.py
  - deduplicate_pptx_media.py
  - index_evidence.py
  - inspect_ppt_media.py
  - prepare_raster_asset.py
  - resource_inventory.py
  - validate_art_direction_plan.py
  - validate_build_deviation_log.py
  - validate_logic_package.py
  - validate_output_qa.py
  - validate_ppt_package.py
  - validate_supervision_report.py
- manifest.yaml
- OFFLINE_DEPENDENCY_NOTICES.md
- THIRD_PARTY_NOTICES.md
- offline-requirements-py312.txt
- v0.4.0-readiness.json
- v0.5.0-readiness.json
- v0.5.1-readiness.json
- v0.5.1-release-notes.md
- v0.5.2-readiness.json
- v0.5.2-release-notes.md
- v0.6.0-readiness.json
- v0.6.0-release-notes.md
- v0.7.0-readiness.json
- v0.7.0-release-notes.md
- v0.7.1-readiness.json
- v0.7.1-release-notes.md
- v0.8.0-readiness.json
- v0.8.0-release-notes.md
- bootstrap_owner_learning.py
- build_provider_manifest.py
- build_runtime_packs.py
- check_release_hygiene.py
- component_version_guard.py
- compose_personal_light.py
- execution_ledger.py
- fetch_offline_wheels.py
- finalize_resource_inventory.py
- index_runtime_cli.py
- install_offline_dependencies.py
- knowledge_cli.py
- materialize_owner_index.py
- migrate_knowledge_index.py
- prepare_release.py
- publish_supervised_pair.py
- run_retrieval_benchmark.py
- runtime_preflight.py
- stamp_pptx_metadata.py
- validate_all.py
- validate_architecture_research_library.py
- validate_composite_skill_mount.py
- validate_config.py
- validate_experience.py
- validate_feedback_benchmark.py
- validate_index_foundation.py
- validate_index_regression_gates.py
- validate_knowledge_scaffold.py
- validate_layout_contracts.py
- validate_markdown_links.py
- validate_pattern_library.py
- validate_personal_extension.py
- validate_personal_extension_foundation.py
- validate_plugin_mount.py
- validate_provenance.py
- validate_resource_inventory_regression.py
- validate_runtime.py
- validate_skill_tree.py
- validate_version.py
- validate_visual_regression_suite.py
- verify_release_bundles.py
- - SKILL.md
- - SKILL.md
- - SKILL.md
- - SKILL.md
- - SKILL.md
- - synthetic-art-direction-plan.json
  - synthetic-copy-package.json
  - visual-regression-suite.json
- test_capability_routing.py
- test_core.py
- test_feedback_benchmark.py
- test_index_runtime.py
- test_layout_contracts.py
- test_pattern_library.py
- test_personal_extension.py
- test_runtime.py
- test_supervision_accountability.py
- test_version_learning.py
.gitattributes
.gitignore
CHANGELOG.md
CITATION.cff
CONTRIBUTING.md
LICENSE
NOTICE
README.md
README.zh-CN.md
requirements.txt
SECURITY.md
VERSION

# Use via CDN

jsDelivr

jsDelivr serves any public GitHub repository as a CDN with zero setup. Pick a version and a file to get a ready-to-paste link and snippet.

Version

File

Link

Example

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

[English](README.md) | [简体中文](README.zh-CN.md)

[![CI](https://github.com/clayzwang/clayz-presentation-skills/actions/workflows/ci.yml/badge.svg)](https://github.com/clayzwang/clayz-presentation-skills/actions/workflows/ci.yml) · Current release: **v0.8.0**

**Compress complex material into presentation-ready decisions — with logic, copy, art direction, editable output, and final QA in one governed workflow.**

## Why Clayz stands out

| Core advantage | What it enables |
| --- | --- |
| **1. Clear architecture** | Five separated stages, shared contracts, centralized configuration, and explicit QA boundaries make the system understandable, replaceable, and auditable. |
| **2. Strong self-learning** | Human-admitted, source-traceable learning can quickly absorb proven methods from many schools of thought without turning reference material into blind imitation. |
| **3. High adaptability** | Task-specific requirements, content constraints, corporate themes, templates, and assets can be composed quickly into one executable production plan instead of being forced into a fixed template. |

![Clayz Presentation Skills workflow](https://raw.githubusercontent.com/clayzwang/clayz-presentation-skills/main/assets/showcase/clayz-overview.png)

## Built for real presentation work

The same system can adapt its visual language to the task while preserving hierarchy, evidence, and editability.

![McKinsey-style consulting presentation example](https://raw.githubusercontent.com/clayzwang/clayz-presentation-skills/main/assets/showcase/mckinsey-demo.png)

![BCG-style consulting presentation example](https://raw.githubusercontent.com/clayzwang/clayz-presentation-skills/main/assets/showcase/bcg-demo.png)

> Showcase slides demonstrate presentation-generation capability and visual adaptation. Brand names shown in examples belong to their respective owners and do not imply affiliation or endorsement.

**[Explore the interactive Experience Center →](https://clayzwang.github.io/clayz-presentation-skills/)**

Clayz Presentation Skills is an open, five-stage system for producing evidence-based presentations:

1. **Logic** establishes the question chain, claims, evidence, and cross-slide invariants.
2. **Copy** locks every visible word, number, break, and atomic copy unit.
3. **Art Direction** turns approved copy into a visual plan without changing content.
4. **Output** builds an editable presentation against the approved plan.
5. **Supervisor** audits drift and evidence without silently redesigning the deck.

## Capability upgrade in this iteration

This is not a larger pile of layout rules. It gives Art Direction a stronger repertoire while keeping every move accountable to the current content:

- **Inventory before authoring:** Supervisor scans the complete plugin runtime, task inputs, owner Library, public Index, brand assets, host capabilities, and font environment before Logic. It tells the user what was found, what will be used, what is unavailable, and which authoring/render route is locked—then authoring may start.
- **Latest components first:** Supervisor checks the official GitHub Latest Release and prints the Public Core, config, runtime, and five-stage contract table. A stale or missing component, an old report, or unavailable freshness evidence fails runtime preflight.
- **One private learning run per version:** the first run reads admitted private knowledge, templates, standards, and methods, builds a persistent index, exercises retrieval probes, and emits a separate learning audit. Later tasks verify hashes and reuse it; same-version source drift never overwrites the first audit silently.
- **First-class Index execution:** one Provider snapshot is locked for the run. Owner learning is materialized from a task-supplied manifest and each stage must attach finalized Retrieval Receipts proving which registered records were actually selected; “reviewed the library” is not evidence.
- **Substance and object gates:** deck-level validators reject mechanical wording and grammar repetition, generic first visuals, unjustified silhouette or structure reuse, shape-only pseudo-charts, missing connectors, and duplicated slide-agnostic QA boilerplate.
- **Content-aware composition:** image-led slides must inspect subject protection, usable copy zones, crop, contrast, directional flow, and overlay necessity before placing text. A remembered advertising layout is never accepted as the answer by itself.
- **Template and icon grammar:** templates, charts, tables, icons, and sample decks enter as reviewed candidates. The plan must re-derive the composition from Logic and Copy, explain every selected asset's semantic role, keep icon-family decisions coherent, and record source and license evidence.
- **Reference-architecture synthesis:** Art Direction ships a validated index of 76 distinct official, diagram-bearing documents from ten publishers and a bilingual 16-card relationship-pattern library. An eight-step method moves from problem framing and source selection through relationship extraction, pattern combination, task-specific derivation, slide translation, diagnosis, and a traceable research ledger—without importing a vendor master.
- **Flexible judgment under contract:** the Art Direction plan v1.4 fixes the evidence and decisions that must be explicit, not coordinates, icon counts, fill ratios, or aesthetic scores. Rendered A/B evidence and professional judgment remain decisive.
- **Fixed regression:** eleven synthetic cases cover four material routes, four cross-slide behaviors, content-aware image composition, governed asset selection, and reference-architecture houses. Tests reject image-led pages without canvas evidence, selected assets without license records, and noun-only houses without responsibility or feedback.
- **Index-native composition methods:** v0.4.0 adds registered Composition and Failure Patterns plus dataset-ready Reference and Sequence metadata. Pattern selection, linked failure knowledge, rejected alternatives, constraints, and expected visual effect are receipt-bound; no match remains explicitly unresolved.
- **Metadata-only dataset boundary:** the public exporter contains stable IDs, classifications, source revisions, hashes, and original method metadata only. It excludes source text, images, coordinates, fonts, templates, corporate assets, model weights, generated-artifact auto-admission, and automatic aesthetic truth.
- **Human-admitted feedback and benchmark:** v0.4.0 re-indexes an observation only when a separate human decision binds its exact hash. Private learning stays outside the public catalog; fixed provider snapshots, four retrieval cases, fail-closed legacy migration, and an explicit release-authorization record turn drift, invention, and unauthorized release into validation failures.
- **Expressive Node.js chart evidence:** v0.5.0 adds an isolated Experience Center route using Apache ECharts SSR and sharp for Sankey and 100% radial stacked-bar assets. The public material includes source data, method notes, reproducible code, web previews, and an editable two-slide PPTX without enabling the security-blocked PptxGenJS adapter.

The public repository still bundles no third-party template, master, icon set, poster, dataset, model, or media. The capability is to retrieve, judge, re-compose, and attribute admissible material—not to clone a source library.

## Project lineage

The five-stage workflow and its first five core capabilities were independently designed and implemented under the **clayz** brand before the project reviewed PPT Master or the other sources listed below. After that foundation was working, Clayz entered an open-learning phase and selectively integrated ideas from multiple open-source and research projects.

[PPT Master](https://github.com/hugohe3/ppt-master) is therefore a later-stage reference—not the origin, fork base, or foundation of this repository. File-level attribution remains governed by the exact provenance and redistribution records under `provenance/`.

## Design principles

- Keep reasoning and visual production as separate approval stages.
- Treat models as capable decision makers; use contracts and scripts only where determinism matters.
- Keep themes, fonts, renderers, reference providers, delivery profiles, and attribution in one central configuration file.
- Never let automated scores replace business, editorial, or visual judgment.
- Preserve editability, source traceability, and final rendered evidence.

## Personal Extension Profile foundation

The v0.5.2 foundation added one optional, owner-private extension decision before Logic while preserving the same five-stage workflow. A private Profile may supply governed theme overrides and logical Library mounts; a private Provider keeps admitted references, indexes, masters, and other assets outside this public repository. Both local and ChatGPT personal hosts resolve the same `library://` URIs through host-specific bindings.

The repository is more than a light archive: one Public Core, the canonical public Provider manifest/index, public Library payloads, shared contracts, and host adapters produce separate Cloud Public Light and Local Public Light targets. The cloud target attaches a resolved Personal Extension Profile and private Library routes to Cloud Public Light; it is not a parallel set of private skills. See [`docs/personal-extension.md`](docs/personal-extension.md) and [`docs/chatgpt-personal-light.md`](docs/chatgpt-personal-light.md). Public material remains usable now. What is deferred is the method for its continuous learning, community aggregation, automatic update, and cross-source fusion, together with Task Overlay, remote MCP Providers, and automatic private Library ingestion.

This public release contains the extension contracts and composition method only. It contains no real owner Profile, private Provider, private index, organization master, font, presentation, PDF, or prior-learning body. v0.8.0 is Windows-first for local execution; Cloud Public Light remains host-managed. No macOS, Linux, or iOS release package is produced or claimed as validated in this version.

On an owner-personal run, Supervisor creates the owner-learning source manifest from discovered resources and stores the real materialization in a repository-external, version-bound private state. The manifest, source bytes, persistent index, and learning audit remain outside public source and release bundles; only the generic materialization/one-time-learning contracts, receipt gates, and synthetic regression fixtures are public.

## Repository layout

- `skills/` contains the five focused skills.
- `catalog/provider-manifest.json` and `catalog/records.jsonl` are the canonical public Provider control plane and index.
- `config/default.json` is the single public configuration source.
- `packages/validators/` contains shared deterministic validators.
- `packages/contracts/` contains versioned cross-stage contracts.
- `packages/layout/` contains the original renderer-neutral relative-layout solver.
- `packages/adapters/` contains optional, separately installed renderer adapters.
- `knowledge/` contains the empty portable learning, source, and index scaffold.
- `examples/` contains synthetic examples only.
- `experience/` contains isolated public-output evidence for the GitHub Pages experience; it is never admitted into the knowledge or reference corpus.
- `provenance/` records conceptual influences and redistributed dependencies.
- `VERSION` is the single release-version source of truth; `scripts/validate_version.py` rejects drift across metadata and public surfaces.

Release maintainers should use `python scripts/prepare_release.py X.Y.Z --date YYYY-MM-DD` instead of editing version strings individually. Merging the resulting `VERSION` change triggers a guarded GitHub Release. See [`docs/releasing.md`](docs/releasing.md).

## Quick start

Clone the complete repository so the five skills keep access to their shared configuration, contracts, validators, and empty knowledge scaffold:

```bash
git clone <your-repository-url> clayz-presentation-skills
cd clayz-presentation-skills
python -m pip install -r requirements.txt
python scripts/validate_all.py
```

Register or expose the five directories under `skills/` to your agent host. Start with Logic and pass each approved artifact to the next stage. Preserve the repository-relative paths; copying one skill directory by itself removes its shared configuration and contracts. Do not skip Supervisor for a final deliverable.

Validate the repository:

```bash
python -m pip install -r requirements.txt
python scripts/validate_all.py
```

Resolve the fully synthetic layout-tree fixture:

```bash
python packages/layout/solve_relative_layout.py \
  examples/synthetic-business-review/layout-tree.json \
  /tmp/resolved-layout.json
```

Resolve, compile, and solve the synthetic registered Layout Contract fixture:

```bash
python packages/layout/compile_layout_contract.py \
  examples/synthetic-layout-contract/comparison-request.json \
  examples/synthetic-layout-contract/comparison-instance.json \
  /tmp/layout-compilation.json \
  --receipt-output /tmp/layout-receipt.json \
  --resolution-output /tmp/layout-resolution.json \
  --resolved-output /tmp/layout-coordinates.json
```

The compiler consumes no Theme or Visual Variant input. If no registered
contract matches, it emits an explicit unresolved resolution and refuses to
invent or compile a fallback contract. See [`docs/layout-contracts.md`](docs/layout-contracts.md).

Resolve and compile the synthetic registered Composition Pattern fixture, then
export the public metadata-only dataset:

```bash
python packages/patterns/compile_composition_pattern.py \
  examples/synthetic-pattern-library/comparison-request.json \
  --resolution-output /tmp/pattern-resolution.json \
  --receipt-dir /tmp/pattern-receipts \
  --plan-output /tmp/composition-plan.json

python packages/patterns/export_metadata_dataset.py \
  /tmp/clayz-metadata-dataset.json
```

The Pattern compiler emits no coordinates and consumes no Theme, Visual
Variant, Layout Contract, or Layout Tree. It refuses to compile when the
Pattern or any linked Failure Pattern is unregistered, unretrieved, ambiguous,
rights-ineligible, or hash-stale. See
[`docs/pattern-dataset-library.md`](docs/pattern-dataset-library.md).

Validate the synthetic feedback loop, retrieval benchmark, migration report,
and release-authorization gate:

```bash
python scripts/validate_feedback_benchmark.py
```

The readiness record requires the five evidence gates plus a separate explicit
human decision before merge, immutable Tag, publication, or the Experience
Center current-release update. See
[`docs/feedback-benchmark-release-readiness.md`](docs/feedback-benchmark-release-readiness.md).

The experimental PptxGenJS route is isolated under `packages/adapters/pptxgenjs/`. It is syntax-checked but security-blocked by default because its current upstream dependency chain contains unpatched denial-of-service advisories; see its README before any evaluation.

Stamp non-visible clayz provenance into a generated PPTX:

```bash
python scripts/stamp_pptx_metadata.py deck.pptx --config config/default.json
```

The metadata stamp is documented, removable, and contains no tracking identifier or network behavior.

Remove only the Clayz-owned custom properties while preserving unrelated document metadata:

```bash
python scripts/stamp_pptx_metadata.py deck.pptx --config config/default.json --remove
```

## Runtime expectations

- The core scripts require Python 3.10 or later. The host-model-independent baseline authoring route uses `python-pptx`; Pillow supports raster, rhythm, and size inspection; PyYAML validates machine-readable manifests.
- GitHub Actions tests Python 3.10, 3.11, and 3.12 and starts every public command-line entry point with `--help` before release.
- Run `scripts/runtime_preflight.py` exactly once per production run. It classifies A–D interaction capabilities, scans dependencies, locks one authoring/render route, and records the configured call budgets. Model brands are not part of the classification.
- The validated local production route for v0.8.0 is Windows with one persistent PowerPoint COM process. Cloud Public Light uses inspected host capabilities. Other local operating-system routes remain future compatibility work and are not release claims.
- PDF and Poppler are lazy dependencies: they are needed for PDF page input or a selected LibreOffice PDF-to-PNG render path, not ordinary PPTX authoring.
- Run `python scripts/build_runtime_packs.py --bundle light` to create Cloud Public Light and Local Public Light from the same public-core digest. Cloud Light relies on ChatGPT host tools and omits local adapters and platform packs; Local Light retains them. Neither archive contains third-party wheels, Experience Center cases, presentation files, or showcase media.
- The only CPython 3.12 offline dependency add-on published for v0.8.0 is Windows x86-64. Maintainers fetch reviewed Windows wheels and build the two Light archives plus the Windows add-on; exact installation, hashes, and licensing are documented in [`docs/release-packages.md`](docs/release-packages.md).
- C and D models use the same internal contracts through an adapter; users continue to interact in natural language and never need to author render manifests or preflight JSON.
- MCP is optional. A capable model host can read the skills and call local tools directly; add MCP only when you want a portable interface to remote storage, search, rendering, or another external service.

See [`docs/runtime-architecture.md`](docs/runtime-architecture.md) for the fixed lifecycle, dependency ladder, platform packs, and host-integration contract.

## Language routing

`config/default.json` defaults to `en-US` and supports `zh-CN`. Every deep reference has an English base file such as `references/example.md` and a matching Chinese file such as `references/example.zh-CN.md`. Each skill resolves the task locale first and loads one language only, unless the user explicitly requests translation comparison. Generated deck language remains a task-level choice; it is not forced by the documentation language.

## Knowledge, learning, and reference index

The public repository includes an intentionally empty portable scaffold under `knowledge/`:

- four learning areas for Logic, Copy, Art Direction, and Output;
- one shared source tree organized by file type;
- empty asset and human-admission registries;
- stage navigation that defines retrieval, writeback, and promotion boundaries.

Supervisor has no separate learning silo. It returns reusable observations to the responsible stage. Generated artifacts, automated scores, usage counts, and Supervisor opinions never become approved references automatically.

Downloading the repository does not create, read, or connect a ChatGPT Library. `catalog/records.jsonl` is the canonical bundled public index. The filesystem `knowledge/` scaffold is an owner-local authoring area; `knowledge/index/search-cache.json`, when generated, is only a derived local search cache and never competes with a Provider index as source of truth. A host binding may map an owner-private Library to ChatGPT without changing its logical `library://` references. See [`knowledge/README.md`](knowledge/README.md) and [`knowledge/registry/schema.md`](knowledge/registry/schema.md).

The scaffold is operational in v0.2.0 and extended by v0.4.0 Stage 5: `scripts/knowledge_cli.py` can register, separately admit, index, search, and record observations. The v2 local index includes learning only after a matching hash-bound human admission. Unadmitted or hash-changed sources are excluded. See [`docs/knowledge-runtime.md`](docs/knowledge-runtime.md).

## Public growth boundary

v0.2.0 adds reusable engineering capability, not someone else's “humanity.” The core skills, synthetic examples, and knowledge scaffold intentionally ship no admitted real cases, taste corpus, personal preference profile, learned corporate style, template, or master. Isolated public-data output evidence may appear under `experience/`, but it is never promoted into the knowledge or reference corpus. Each user may build those layers through the governed knowledge scaffold. The deliberately excluded technologies and the exact adopted/not-adopted boundary are documented in [`docs/source-adoption.md`](docs/source-adoption.md).

## Configuration

Edit or override `config/default.json`. Do not hard-code organization brands, master files, fonts, local paths, or renderer names inside a skill.

The core package intentionally includes no uploaded master, private presentation, PDF, font, or derived corporate reference asset. The example under `examples/` remains fully synthetic. Public-data screenshots and presentations under `experience/` are output evidence only and are governed by `experience/case-manifest.json` plus release-hygiene scanning.

Release validation accepts an optional, untracked denylist through `CLAYZ_RELEASE_DENYLIST` or `scripts/check_release_hygiene.py --denylist`. Keep organization-specific names and source-material phrases in that local file, never in the public repository.

For GitHub CI, store the UTF-8 denylist as a base64-encoded repository secret named `CLAYZ_RELEASE_DENYLIST_B64`. The workflow decodes it only on the runner and never commits the terms. Before a public release, also enable GitHub private vulnerability reporting as described in [`SECURITY.md`](SECURITY.md).

## Influence citations

Later-stage conceptual influences are identified at the exact reference points inside the skill documentation and summarized in `provenance/manifest.yaml`. With thanks to the authors and contributors of [PPT Master](https://github.com/hugohe3/ppt-master), [PPTAgent](https://github.com/icip-cas/PPTAgent), [DeepPresenter](https://arxiv.org/abs/2602.22839), [pom](https://github.com/hirokisakabe/pom), [VASCAR](https://arxiv.org/abs/2412.04237), [PosterO](https://openaccess.thecvf.com/content/CVPR2025/html/Hsu_PosterO_Structuring_Layout_Trees_to_Enable_Language_Models_in_Generalized_CVPR_2025_paper.html), [PosterLayout](https://arxiv.org/abs/2303.15937), [Scan-and-Print](https://arxiv.org/abs/2505.20649), [CreatiPoster](https://arxiv.org/abs/2506.10890), [Tahta](https://github.com/zcag/tahta), [MiniMax Skills](https://github.com/MiniMax-AI/skills), [PosterCopilot](https://github.com/JiazheWei/PosterCopilot), [PosterVerse](https://github.com/wuhaer/PosterVerse), and [PptxGenJS](https://github.com/gitbrent/PptxGenJS). Exact revisions, licenses, citation-only constraints, and non-redistribution boundaries are recorded; no upstream source snapshot, prompt, template, reference slide, media, dataset, or model is bundled.

For the reference-architecture synthesis method, Clayz also thanks the architects, technical writers, designers, engineers, and reviewers behind [IBM Think Architectures](https://www.ibm.com/think/architectures), [Microsoft Azure Architecture Center](https://learn.microsoft.com/en-us/azure/architecture/), [Google Cloud Architecture Center](https://cloud.google.com/architecture), [AWS Architecture Center](https://aws.amazon.com/architecture/), [Oracle Architecture Center](https://docs.oracle.com/solutions/), [SAP Architecture Center](https://architecture.learning.sap.com/), [NVIDIA Enterprise Reference Architectures](https://docs.nvidia.com/enterprise-reference-architectures/), [Databricks reference architectures](https://docs.databricks.com/aws/en/lakehouse-architecture/reference), [Snowflake architecture guidance](https://www.snowflake.com/en/developers/guides), and [Apple Platform Security](https://support.apple.com/guide/security/welcome/web). The index records links and distilled relationships only; no source diagram, wording, icon, brand identity, master, template, coordinate system, or media is redistributed.

## License and citation

Licensed under Apache-2.0. See `NOTICE`, `CITATION.cff`, and `provenance/THIRD_PARTY_NOTICES.md`.

clayz-presentation-skills

File Explorer

# Use via CDN

Was this content helpful?

Recent Feedback

Download README