WhiteKiwi Skills

Portable Agent Skills for Claude Code, Codex and ChatGPT, and OpenClaw. Each workflow is published as a separately installable plugin or skill, so you can install only what you need.

Available skills

Skill	What it does	Requires	Install name
Locron	Safely operate and diagnose local schedules	locron on PATH	locron
Pushman	Safely send and inspect personal iPhone notifications	pushman on PATH	pushman

The marketplace is the catalog, not an all-in-one bundle. Adding it makes the entries discoverable; it does not install every plugin. Install one or both entries explicitly.

Install only what you need

Claude Code

Register the catalog once:

claude plugin marketplace add WhiteKiwi/skills

Then choose a plugin:

claude plugin install locron@whitekiwi-skills
# or
claude plugin install pushman@whitekiwi-skills

Codex and ChatGPT

Register the catalog once:

codex plugin marketplace add WhiteKiwi/skills

Then choose a plugin:

codex plugin add locron@whitekiwi-skills
# or
codex plugin add pushman@whitekiwi-skills

Codex CLI uses plugin add for installation. The same catalog is available in the ChatGPT desktop Plugins directory after registration, where Locron and Pushman remain separate install choices.

OpenClaw

Install a skill directly by its owner-qualified ClawHub reference:

openclaw skills install @whitekiwi/locron
# or
openclaw skills install @whitekiwi/pushman

The owner-qualified ClawHub reference is the supported registry path. Review the current registry scan before installing. For local development, use the generated payload described below.

Use the Locron skill

Ask the agent to use Locron, or invoke the skill explicitly on clients that expose named skill invocation. For example:

Use the locron skill to explain why the backup job did not run.

The workflow follows four operating rules:

• discover the installed Locron version and help surface before composing commands;
• prefer versioned locron.cli/v1 JSON for observations and decisions;
• dry-run supported mutations, inspect the normalized result, then apply only when the request authorizes it;
• finish mutations with an exact-target read-back and diagnoses with durable evidence.

The workflow is tested through Locron 0.8.0. It uses explain for the preferred consolidated job report, why --run for the full immutable attempt and event trace, and the installed dashboard help surface for safe local dashboard lifecycle operations. It falls back to the capabilities exposed by older installed versions instead of assuming newer commands exist.

Read the authored workflow in skills/locron/SKILL.md and its on-demand safety reference.

Use the Pushman skill

Install Pushman CLI 0.1.0 or newer, authorize it through a browser or the iPhone app, and optionally connect its local stdio MCP server:

pushman login
# or: pushman pair
codex mcp add pushman -- pushman mcp

Then ask the agent directly:

Use the pushman skill to notify me when this task finishes.

The workflow prefers typed MCP tools, falls back to the installed CLI, treats a direct exact-send request as authorization for one send, and never turns a draft or inspection request into a notification. It does not automatically retry ambiguous results because a retry can create a duplicate and consume quota.

Read the authored workflow in skills/pushman/SKILL.md and its on-demand safety reference.

Update or remove

Claude Code

claude plugin marketplace update whitekiwi-skills
claude plugin update locron@whitekiwi-skills
claude plugin update pushman@whitekiwi-skills
claude plugin uninstall locron@whitekiwi-skills
claude plugin uninstall pushman@whitekiwi-skills

Codex and ChatGPT

codex plugin marketplace upgrade whitekiwi-skills
codex plugin remove locron@whitekiwi-skills
codex plugin remove pushman@whitekiwi-skills
codex plugin add locron@whitekiwi-skills
codex plugin add pushman@whitekiwi-skills

Remove the catalog itself only when it is no longer needed:

codex plugin marketplace remove whitekiwi-skills

OpenClaw

openclaw skills update @whitekiwi/locron
openclaw skills update @whitekiwi/pushman

The current native OpenClaw CLI does not expose skills uninstall. The standalone clawhub uninstall command applies to installations tracked by the standalone ClawHub CLI, not automatically to native OpenClaw-managed installations.

Distribution model

The install boundary is a plugin, not the entire repository. WhiteKiwi publishes an independent plugin for each standalone workflow. A plugin may contain more than one skill only when those skills form one coherent capability that users would normally install together.

Client	Distribution	Generated metadata
Claude Code	Git-hosted plugin marketplace	.claude-plugin/plugin.json
Codex and ChatGPT	Git-backed plugin marketplace	.codex-plugin/plugin.json and agents/openai.yaml
OpenClaw	ClawHub Agent Skill	metadata.openclaw.requires.bins

Each skills/<name>/ directory is an authored workflow. Its generated plugins/<name>/ package contains only that workflow, so installing locron does not install pushman, and vice versa. scripts/build.sh produces every adapter from those sources, injects only platform-specific metadata, and rejects generated drift. Release archives are deterministic and include checksums.

Local development

Clone and validate the repository:

git clone https://github.com/WhiteKiwi/skills.git
cd skills
./scripts/validate.sh

Use a local package without publishing it:

claude plugin marketplace add .
codex plugin marketplace add .
openclaw skills install ./platforms/openclaw/locron --as locron
openclaw skills install ./platforms/openclaw/pushman --as pushman

Build all generated payloads and reproducible release archives:

./scripts/build.sh
./scripts/validate.sh

dist/
├── claude/<skill>/
├── openclaw/<skill>/
├── codex/<skill>/
├── skill/<skill>/
├── <skill>-claude-<version>.zip
├── <skill>-openclaw-<version>.zip
├── <skill>-codex-<version>.zip
├── <skill>-skill-<version>.zip
└── SHA256SUMS

Each <skill>-skill-<version>.zip contains only that portable skill. These archives are release and API convenience artifacts, not documented ChatGPT installation ZIPs.

Repository layout

skills/<name>/                         authored sources of truth
plugins/<name>/                        generated, independently installable plugin
.claude-plugin/marketplace.json        generated catalog of Claude plugins
.agents/plugins/marketplace.json       generated catalog of Codex/ChatGPT plugins
platforms/openclaw/<name>/             generated OpenClaw skills
catalog.json                           catalog and packaging metadata
scripts/                               deterministic build, validation, release
tests/                                 trigger, packaging, and behavior tests
.github/workflows/                     validation and tagged release

Do not edit generated files directly. Change skills/<name>/, catalog.json, or VERSION, then run ./scripts/build.sh.

Release

Maintainers update VERSION, rebuild, validate, and create the matching tag:

./scripts/publish.sh --dry-run
git tag "v$(cat VERSION)"
git push origin main --tags

The v* workflow verifies that the tag matches VERSION, creates an idempotent GitHub Release, and attaches every ZIP plus SHA256SUMS. When CLAWHUB_TOKEN is configured, it also publishes the generated OpenClaw payload with the same explicit version.

For an authenticated manual ClawHub publication:

./scripts/publish.sh --clawhub

References

• Agent Skills specification
• Agent Skills authoring best practices
• OpenAI plugin packaging
• OpenAI plugin examples
• Claude Code plugin marketplaces
• OpenClaw skills and ClawHub

License

This catalog and its skill artifacts are licensed under MIT-0. Locron retains its own licenses; no Locron implementation code is included here.