ArkSpace

中文

ArkSpace is a host-neutral workspace for reusable agent skills, callable roles, and workflow protocols. It gives an AI agent focused instructions for the task at hand while keeping sources canonical, providers explicit, and host adapters generated.

Start Here

Install the published skills into the current supported host:

# Install all discovered ArkSpace skills in this project
npx skills add arch3rPro/ark-space

# Install one skill
npx skills add arch3rPro/ark-space --skill web-search

# Install one skill globally
npx skills add arch3rPro/ark-space --skill web-search -g

Use --agent <host> to target a specific supported host. From a local checkout, run npx skills add ..

Then choose the shortest path to a useful result:

# Let ArkSpace choose the role, workflow, capability, and provider
/ark-space:orchestrator research the AI coding agents market

# Invoke a known capability directly
/ark-space:web-search search diffusion transformers
/ark-space:web-fetch extract https://example.com/article

See docs/invocation.md for the complete direct and Orchestrator invocation contract.

How It Works

ArkSpace keeps four host-neutral layers:

• skills/ — focused, callable instructions for a capability.
• agents/ — role-specific behavior profiles for hosts that support callable agents.
• workflows/ — routing, quality, provider, and handoff protocols.
• registry/ — public inventory, source governance, and capability-specific provider policy.

The Orchestrator classifies an ambiguous request and selects the smallest useful role, workflow, and capability. Capability selection precedes provider selection; a user-requested provider is never silently replaced. Claude Code, Codex, and future hosts consume the same canonical skills through adapters.

Capabilities

Route, Govern, and Configure

• orchestrator routes tasks across roles, workflows, capabilities, and providers.
• skill-manager creates, adapts, validates, and source-tracks skills.
• provider-manager configures providers, key references, readiness, and rotation.

Search, Read, Research, and Automate

• web-search, web-fetch, web-site, web-research, web-extract, and web-automation cover discovery, reading, mapping, cited synthesis, structured extraction, and browser work.
• defuddle extracts clean Markdown from standard web pages.
• code-context retrieves implementation-oriented examples and API context.

Plan, Organize, and Execute

• prd-create produces screenshot-first R&D PRDs.
• drive-me turns stalled personal work into a bounded next-action plan.
• json-canvas, obsidian-markdown, obsidian-bases, obsidian-kanban, and obsidian-cli support structured knowledge work in Obsidian.

Included Skills

Skill	Purpose
orchestrator	Route a request to the smallest useful role, workflow, capability, and provider.
skill-manager	Create, adapt, validate, and source-track skills.
provider-manager	Configure providers, readiness, and key rotation.
defuddle	Extract clean Markdown from standard web pages.
web-search	Discover public sources and related pages.
web-fetch	Extract content from a supplied URL.
web-site	Map or crawl a requested site section.
web-research	Produce cited research.
web-extract	Extract schema-shaped public data.
web-automation	Interact with live pages or manage monitors.
code-context	Retrieve implementation examples and API context.
json-canvas	Create and edit JSON Canvas files.
obsidian-bases	Create and edit Obsidian Bases.
obsidian-cli	Use the Obsidian CLI for app-level operations.
obsidian-kanban	Create and maintain Obsidian Kanban boards.
obsidian-markdown	Create and edit Obsidian Flavored Markdown.
drive-me	Turn stalled work into a bounded next-action plan.
prd-create	Create screenshot-first R&D PRDs.

Callable Roles

Use role-specific behavior when the host supports callable agents:

• Code: code-engineer, code-reviewer
• Documentation and research: doc-writer, web-researcher, knowledge-manager
• Product and project: prd-planner, competitive-analyst, project-manager
• Personal and governance: personal-assistant, skill-manager, orchestrator

Providers Stay Explicit

Provider-backed skills use capability registries and local configuration. Secrets and private endpoints stay outside the committed package, and missing configuration is reported instead of being hidden by an unsupported fallback.

Start with docs/provider-configuration.md for URLs, API keys, multi-key rotation, setup recovery, and guided commands.

Project Map

Need	Read
Choose direct invocation or Orchestrator routing	docs/invocation.md
Understand layers and runtime entry points	docs/architecture.md
Configure providers	docs/provider-configuration.md
Add or adapt a skill	docs/adding-skills.md
Check host support	docs/platform-support.md
Maintain, package, and validate ArkSpace	docs/maintenance.md
Review planned work and release checks	docs/roadmap.md

Development Contract

• Canonical skills live in skills/<skill-name>/SKILL.md.
• Callable agent sources live in agents/; orchestration protocols live in workflows/.
• Provider and source governance live in registry/.
• integrations/ and plugins/ark-space/ are generated outputs.
• Private configuration does not belong in the public package.

For maintainer validation and packaging commands, use docs/maintenance.md.