AI format catalog
Bundle OKF 0.1 · 3 conceitos · xiuuu4499/ai-plugin-marketplace
Open source Repository Open in the app JSON README (API)
About
# AI format catalog
* [Platforms](platforms/) - vendor-specific packaging and capability facts
* [Components](components/) - reusable artifact semantics independent of vendor
* [Mappings](mappings/) - equivalence, approximation, and fidelity policies
Details
- Kind
- OKF bundles
- Topic
- No topic detected
- Publisher
- xiuuu4499
- Origin
- okf_github
- Category
- dados
- Version
- 0.1
- Last push
- 2026-08-21T01:01:32Z
- Repository state
- ativo
- Language
- Python
- License
- MIT
- Added
- 2026-09-08 02:26:02
- Updated
- 2026-09-08 02:26:02
- Origin id
xiuuu4499/ai-plugin-marketplace:plugins/ai-transmute/knowledge/index.md
README
# AI Plugin Marketplace A dedicated Codex marketplace for portable AI-extension tooling. It currently ships six plugins: | Plugin | Purpose | | --- | --- | | `okf` | Author, search, migrate, validate, lint, maintain, and curate [Open Knowledge Format](https://github.com/GoogleCloudPlatform/knowledge-catalog/tree/main/okf) bundles. | | `ai-transmute` | Inspect and convert AI skills, commands, agents, hooks, plugins, instruction files, MCP/LSP configuration, and notebooks through auditable OKF intermediates. | | `ai-scholar` | Research a subject deeply, curate cited OKF knowledge, and design a human-approved Codex plugin. | | `github-docs-api` | Search, retrieve, and build upon GitHub documentation via the official GitHub Docs API. | | `kilo-qwen38-generation-policy` | Install a global Kilo Code plugin that enforces explicit Qwen3.8 request parameters. | | `local-model-workbench` | Plan and troubleshoot llama.cpp, Qwen3.8, MiniMax H3, ComfyUI, VS Code, and local-model workflows. | The marketplace lives at [`.agents/plugins/marketplace.json`](.agents/plugins/marketplace.json). All plugins are maintained in this repository and published by [xiuuu4499](https://github.com/xiuuu4499). ## Scope AI Transmute understands Codex, Claude, GitHub Copilot, Grok Build, and Gemini CLI packaging. Its workflow is deliberately loss-aware: 1. Inventory the source without changing it. 2. Retrieve the relevant platform and component knowledge. 3. Create and curate an OKF audit bundle for the conversion job. 4. Render the target package. 5. Validate the result and report exact mappings, approximations, omissions, and manual follow-ups. Conversions are written to a new timestamped directory and contain: ```text ai-transmute/<timestamp>-<source>-to-<target>/ ├── package/ # converted target artifact ├── okf/ # portable conversion record ├── report.md # fidelity and validation report └── job.json # machine-readable job metadata ``` Jupyter notebooks retain their cells, metadata, attachments, and code. Hosted AI notebooks or projects without a documented portable import API are rendered as import packs with manual setup instructions. AI Scholar uses a separate, evidence-led workflow: 1. Collect official and independent sources, preserving a reviewable record of duplicates and exclusions. 2. Prioritize retained sources, contradictions, uncertainty, and special cases. 3. Build and curate a cited OKF knowledge bundle after each source import. 4. Design a traceable plugin interface and present it for explicit human approval. 5. Implement, validate, and hand off only the approved design. Its stateful job record prevents implementation before the review artifact records explicit approval. AI Scholar’s primary user interface is its `study`, `sources`, `knowledge`, `architect`, `review`, `implement`, `validate`, `maintain`, `doctor`, and `plan-mode` skills; its small Python CLI persists state and verifies gates. The optional curation hook is advisory and fail-open. ## Install Codex CLI in a Codespace OpenAI's current macOS/Linux recommendation is the standalone installer. It installs in user space and does not require rebuilding or restarting the devcontainer: ```bash curl -fsSL https://chatgpt.com/codex/install.sh | sh hash -r codex --version ``` The standard Codespaces `PATH` already includes `~/.local/bin`. If a customized image does not, activate it in the current shell: ```bash export PATH="$HOME/.local/bin:$PATH" hash -r ``` Run `codex` once and select a supported sign-in method if the CLI requests authentication. For reproducible future container builds, put the same installer command in the consuming repository's devcontainer setup. That is optional for the current container and only takes effect on a later create or rebuild. ## Install this marketplace From the root of this clone: ```bash codex plugin marketplace add "$PWD" codex plugin add okf@ai-plugin-marketplace codex plugin add ai-transmute@ai-plugin-marketplace codex plugin add ai-scholar@ai-plugin-marketplace codex plugin add github-docs-api@ai-plugin-marketplace codex plugin add kilo-qwen38-generation-policy@ai-plugin-marketplace codex plugin add local-model-workbench@ai-plugin-marketplace ``` Start a new Codex thread after installation so newly installed skills and hooks are loaded. The advisory hooks are fail-open: they provide capped validation context but do not rewrite files or grant permissions. ## Plugin entry points The `okf` plugin exposes the full OKF workflow plus a thin `gem` command router. AI Transmute exposes these skills: - `inspect` — detect the platform, components, dependencies, and portability hazards without writing output. - `plan` — produce a dry-run mapping and predicted loss report. - `convert` — perform a format-preserving conversion through an OKF job bundle. - `extract` — pull selected components from a package or extrapolate a package from a smaller artifact. - `transform` — redesign behavior while explicitly recording semantic changes. - `learn-format` — curate official specifications and local examples before enabling a format. - `feedback` — correct durable format knowledge, mappings, renderer behavior, or one conversion job. - `validate` — check sources, OKF bundles, target packages, and complete jobs. - `diff` — compare behavior and component inventories rather than filenames alone. - `doctor` — report required and optional runtime availability. AI Scholar exposes these skills: - `study` — initialize a research contract and auditable job. - `sources` — collect, deduplicate, rank, and retain source decisions. - `knowledge` — create and curate cited OKF concepts in source-priority order. - `architect` — design skills, scripts, hooks, and other surfaces from curated evidence. - `review` — present the complete design packet and record explicit approval or requested changes. - `implement` — enforce approval before building the approved plugin. - `validate` — check source provenance, job state, OKF curation, traceability, and package quality. - `maintain` — assess source drift and revise research without overwriting prior evidence. - `doctor` — diagnose missing dependencies and validation failures. - `plan-mode` — map a Codex Plan-mode plan to the current AI Scholar job state and required gates. GitHub Docs API exposes these skills: - `answer-short-question` — quickly look up and answer a focused question using the GitHub Docs Search and Article Body REST APIs. - `extract-github-sme-plugin` — create a new GitHub domain Subject Matter Expert agent plugin by combining GitHub Docs API content extraction with the ai-scholar research and design workflow. - `update-github-apis` — re-crawl the GitHub Docs API surface to detect documentation changes and refresh the plugin's knowledge base. - `write-api-script` — create a standalone shell or Python script that calls GitHub Docs APIs directly to fetch or process documentation content without AI tooling. Kilo Qwen3.8 Generation Policy exposes: - `qwen38-generation-policy` — choose and audit named Qwen3.8 thinking/instruct profiles, with request-time enforcement supplied by the bundled Kilo plugin. Local Model Workbench exposes: - `local-model-workbench` — plan, inspect, document, and troubleshoot local model workflows across Kilo, llama.cpp, llama-swap, ComfyUI, MiniMax H3, MCP, and VS Code. ## Install Kilo integration After installing the Codex plugin from this marketplace, install its Kilo artifacts in the environment where Kilo runs: ```bash plugins/kilo-qwen38-generation-policy/scripts/install-kilo-qwen38-policy.sh ``` PowerShell: ```powershell .\plugins\kilo-qwen38-generation-policy\scripts\install-kilo-qwen38-policy.ps1 ``` The installer writes only owned files under `~/.config/kilo` by default and can be redirected with `KILO_TARGET_CONFIG_DIR`. The native implementation is Python; `.sh` and `.ps1` wrappers are provided for platform-native invocation. Run `/reload` in Kilo or restart Kilo after installation. Uninstall: ```bash plugins/kilo-qwen38-generation-policy/scripts/uninstall-kilo-qwen38-policy.sh ``` ## Requirements - Python 3 for AI Transmute's deterministic engine and advisory hook. - Python 3 for AI Scholar's deterministic job engine and advisory hook. - Python 3 for the GitHub Docs API plugin's validation script. - The `okf` CLI for audited conversion and knowledge-catalog workflows. - Ruby for the OKF plugin's advisory curation hook. - Target-native CLIs are optional; when present, validation can use them in addition to internal checks. Missing optional tools are reported rather than installed automatically. ## Development and validation Run the deterministic test suite: ```bash python3 -m unittest discover -s plugins/ai-transmute/tests -v python3 -m unittest discover -s plugins/ai-scholar/tests -v python3 -m unittest discover -s plugins/github-docs-api/tests -v python3 -m unittest discover -s plugins/kilo-qwen38-generation-policy/tests -v ``` Validate the persistent knowledge catalog: ```bash okf validate plugins/ai-transmute/knowledge okf lint plugins/ai-transmute/knowledge okf loose plugins/ai-transmute/knowledge ``` Before committing an AI Scholar plugin change, also run: ```bash python3 plugins/ai-scholar/scripts/ai_scholar.py init "smoke test" --output /tmp --json python3 /home/codespace/.codex/skills/.system/plugin-creator/scripts/validate_plugin.py plugins/ai-scholar ``` Plugin manifests and skill folders can additionally be checked with the validators bundled with Codex's `plugin-creator` and `skill-creator` skills. ## License Repository-authored code and documentation are available under the [MIT License](LICENSE). Vendored OKF reference material retains its upstream Apache 2.0 notice in [`plugins/okf/skills/okf/reference/APACHE-2.0.txt`](plugins/okf/skills/okf/reference/APACHE-2.0.txt).