STATUS: v1.4.1. Cross-agent skill for authoring, converting, and validating Open Knowledge Format (OKF) Markdown — installable in Claude Code and Codex. Repo: https://github.com/parkscloud/okf-author · Design + decision log:
DESIGN.md.
A cross-agent Agent Skill that helps users author Markdown in Open Knowledge Format (OKF) — and review/convert existing Markdown into it — from inside Claude Code and Codex. Open-source (MIT) and public: anyone can install it.
OKF is an open, vendor-neutral specification for representing knowledge — the metadata,
context, and curated insight around your data and systems — as a directory of Markdown
files with YAML frontmatter. Each file is one concept; files interlink with ordinary
Markdown links; the only required frontmatter field is type (with title, description,
resource, tags, and timestamp recommended).
- Introduced: 2026-06-12 by Google Cloud, released as OKF v0.1 — formalizing the emerging "LLM-wiki" pattern into a portable, interoperable standard.
- License: the OKF specification itself is Apache-2.0.
- Authoritative links:
- Specification: https://github.com/GoogleCloudPlatform/knowledge-catalog/blob/main/okf/SPEC.md
- Announcement (Google Cloud blog): https://cloud.google.com/blog/products/data-analytics/how-the-open-knowledge-format-can-improve-data-sharing
- Reference repository: https://github.com/GoogleCloudPlatform/knowledge-catalog
- No lock-in. It's just Markdown files in a directory — no proprietary platform,
database, SDK, or account. If you can
cata file you can read it; if you cangit clonea repo you can ship it. - Readable by humans and agents. It renders on GitHub, opens in any editor, and is indexable by any search tool — while giving AI agents the accurate, curated context they need to reason about your data and systems.
- Durable and diffable. One version-controlled home for your knowledge, living next to the code it describes; every change shows up in a normal diff.
- Interoperable. A small set of agreed conventions lets knowledge written by one producer be consumed by different agents and tools without translation.
okf-author is one skill with three modes:
- Author — new Markdown comes out OKF-conformant: a
typeplus smart-default frontmatter (title,description,timestamp, andtags/resourcewhen applicable). - Convert — bring existing Markdown into OKF safely and in stages (frontmatter first,
structure second); in place when the directory is a clean git repo, otherwise into a
parallel copy. Non-destructive by default. The bundled
generate_indexes.pybuilds the per-folderindex.md/log.mddeterministically. - Validate — check a bundle against OKF v0.1 §9 conformance with the bundled,
dependency-free
validate.py.
It activates when you mention OKF or work inside an existing OKF bundle, and otherwise
offers OKF once when you're authoring substantial Markdown. Full behavior:
skills/okf-author/SKILL.md.
Install from this repo's built-in marketplace:
/plugin marketplace add parkscloud/okf-author
/plugin install okf-author@okf-author
Turn on auto-update — it's off by default for this marketplace. Self-hosted marketplaces
don't update on their own, so you have to opt in: open the /plugin interface and go to
Marketplaces → okf-author → Enable auto-update. Claude Code then refreshes the plugin at
the start of each new session (start a fresh session to pick up a new version). If you skip this
step, the plugin stays pinned at the version you installed until you update it by hand —
claude plugin marketplace update, then claude plugin update okf-author@okf-author. (Codex
has no plugin system — use the script install below.)
From a clone of this repo (Python 3, standard library only — no pip install):
python3 install.py --all # install into Claude Code and Codex
python3 install.py --claude # Claude Code only
python3 install.py --codex # Codex only
python3 install.py --all --dry-run # preview, change nothing(On Windows use python or py instead of python3.)
It installs the skill folder into:
- Claude Code:
~/.claude/skills/okf-author/ - Codex:
~/.agents/skills/okf-author/(per OpenAI's Codex skills docs; override with--codex-dir)
Then restart your agent (or open a new session) so it discovers the skill, and mention OKF or ask it to author / convert / validate Markdown.
Both bundled tools are dependency-free (PyYAML used only if already installed) and run on their own:
# validate a bundle against OKF v0.1 conformance
python3 skills/okf-author/validate.py examples/handbook # -> CONFORMANT
python3 skills/okf-author/validate.py --strict path/to/bundle # also require title/description/timestamp
python3 skills/okf-author/validate.py --json path/to/bundle # machine-readable
# (re)generate index.md + log.md across a bundle from its frontmatter
python3 skills/okf-author/generate_indexes.py path/to/bundle --title "My Knowledge Base"
python3 skills/okf-author/generate_indexes.py path/to/bundle --dry-runExit code 0 = conformant, 1 = errors, 2 = bad path. Warnings (missing recommended
fields, non-ISO timestamps, broken links) are advisory and never fail a bundle. PyYAML is
used if installed; otherwise a built-in minimal parser handles the conformance checks.
okf-author/
├── .claude-plugin/ # plugin.json + marketplace.json — Claude Code plugin + marketplace
├── README.md · DESIGN.md · CLAUDE.md · LICENSE · install.py
├── skills/okf-author/ # the skill — installed by install.py and bundled into the plugin
│ ├── SKILL.md # the skill: Author / Convert / Validate
│ ├── validate.py # dependency-free OKF v0.1 conformance checker
│ ├── generate_indexes.py # deterministic index.md / log.md generator
│ └── reference/ # vendored OKF spec (verbatim) + license + attribution
└── examples/handbook/ # a tiny conformant example bundle
Created and maintained by Robert Parks — raparks@icloud.com. Issues and contributions welcome.
Licensed MIT for okf-author's own code. The vendored OKF specification
(skills/okf-author/reference/SPEC.md) remains © Google LLC under Apache-2.0 and is
included verbatim with attribution (see skills/okf-author/reference/).
- Agent Skills (Anthropic, open
SKILL.mdformat): https://github.com/anthropics/skills - Codex skills: https://developers.openai.com/codex/skills
- Prior-art community OKF skill (reference only): https://github.com/scaccogatto/okf-skills
- OKF specification and announcement: see About Open Knowledge Format (OKF) above.