ArkSpace

中文 | English

ArkSpace is an Agent Skills workspace for Claude Code, Codex, and compatible AI-agent hosts. It packages reusable skills, callable agent roles, workflow protocols, provider routing metadata, and host adapters so an AI agent can select focused, reusable context for each task.

Skills are the public contract. Runtime scripts and provider CLIs support skills when a skill needs configuration, search, extraction, or validation.

How Agents Use ArkSpace

Use ArkSpace from an AI-agent session with slash invocation.

/ark-space:orchestrator search for the claude-code-everything project
/ark-space:web-search search diffusion transformers
/ark-space:web-search search Claude Code plugin docs
/ark-space:web-research research the AI coding agents market
/ark-space:web-fetch extract https://example.com

Choose the entry path by intent:

Path	Use When
/ark-space:orchestrator ...	You want ArkSpace to choose the role, workflow, capability, and provider.
/ark-space:<skill-name> ...	You know the task capability; select a provider only when its distinctive behavior matters.
agents/*	The host supports callable agents/subagents and should use a role-specific behavior profile.

See docs/invocation.md for the full invocation contract and capability split.

Personal execution examples:

/ark-space:orchestrator help me run my weekly planning board
/ark-space:orchestrator capture these personal tasks into my Obsidian Kanban

Core Model

ArkSpace has four host-neutral layers:

Layer	Purpose
skills/	Canonical Agent Skills. Each public skill lives at skills/<name>/SKILL.md.
agents/	Callable role definitions that compose skills and workflows without duplicating skill bodies.
workflows/	Routing, handoff, provider selection, and quality-gate protocols.
registry/	Source governance, role ownership, skill inventory, provider metadata, and validation contracts.

Host-specific files are adapters:

Path	Purpose
.claude-plugin/	Claude Code plugin metadata.
.codex-plugin/	Codex plugin metadata.
integrations/	Generated host-native agent outputs.
plugins/ark-space/	Packaged Codex marketplace copy generated from canonical sources.

Claude Code, Codex, and future hosts consume the same skill files through adapters.

Callable Agents

Agent	Owns
orchestrator	Lightweight routing, provider setup routing, workflow selection.
code-engineer	Implementation, refactoring, tests, debugging.
code-reviewer	Bug, regression, risk, and test-gap review.
doc-writer	Project documentation and Obsidian-flavored Markdown when needed.
web-researcher	Web search, URL extraction, crawling, monitors, and source-grounded research.
knowledge-manager	Notes, Obsidian artifacts, Bases, Canvas, Kanban, and vault organization.
prd-planner	Requirements, scope, acceptance criteria, product decisions.
competitive-analyst	Product, competitor, market, and public-evidence analysis, with web operations handed to the web researcher when needed.
project-manager	Milestones, task breakdown, risks, status structures.
personal-assistant	Personal task capture, weekly planning, Kanban-first personal execution, and personal project upkeep.
skill-manager	Skill lifecycle, upstream provenance, registries, package integrity.

Included Skills

Core And Governance

Skill	Purpose
orchestrator	Route work to the smallest useful role, workflow, capability, and provider.
skill-manager	Create, adapt, validate, source-track, and govern ArkSpace skills.
provider-manager	Configure and inspect provider URLs, key references, readiness, and rotation.

Personal Execution

Skill	Purpose
drive-me	Turn personal execution friction, scope drift, or stalled work into one bounded next-action plan.

Product Requirements

Skill	Purpose
prd-create	Keep standard R&D PRD sections and functional-page fields/constraints while writing workflows as concise screenshot-first requirements.

Search, Fetch, And Research

Skill	Purpose
web-search	Search public sources, papers, or self-hosted metasearch; or find pages related to a known URL. Zero-config default is Exa MCP; also supports Exa, Tavily, Firecrawl, Jina, DuckDuckGo, Brave, SearXNG, and arXiv.
defuddle	Extract clean Markdown from normal web pages through Defuddle CLI.
web-fetch	Extract URL content through Exa, Tavily, or Firecrawl.
web-site	Map a site or crawl a requested site section.
web-research	Produce cited research through Exa or Tavily.
web-extract	Extract schema-shaped public data through Firecrawl.
web-automation	Interact with live pages or manage recurring monitors.
code-context	Retrieve implementation-oriented examples and API usage context through Exa.

Provider-specific implementations are internal adapters. Public skills remain capability-based; select an explicit mode and pass a provider only when a user requests it or its distinctive controls are required.

Knowledge And Obsidian Tools

Skill	Purpose
json-canvas	Create and edit JSON Canvas files.
obsidian-bases	Create and edit Obsidian Bases.
obsidian-cli	Interact with Obsidian through the CLI.
obsidian-kanban	Create and maintain Obsidian Kanban boards.
obsidian-markdown	Create and edit Obsidian-flavored Markdown.

The Obsidian skills are retained as knowledge-management tools within the broader ArkSpace package.

Provider Configuration

ArkSpace keeps private provider configuration outside the public repository. When a provider is missing a URL or API key, the skill should guide the user through setup with local configuration and private secrets.

Provider setup supports:

• self-hosted service URLs such as SearXNG
• API-backed providers such as Exa, Tavily, Firecrawl, and Brave
• zero-config keyless providers such as Exa MCP, Jina, and DuckDuckGo (no setup needed)
• multiple API keys with rotation
• local private secret storage or environment-variable references

See docs/provider-configuration.md for command setup, manual recovery, and agent-guided setup.

Documentation

Document	Purpose
docs/invocation.md	Slash invocation, direct skills, Orchestrator routing, capability split.
docs/provider-configuration.md	Provider URLs, API keys, multi-key rotation, setup recovery.
docs/roadmap.md	Project roadmap, development workstreams, and release readiness checks.
docs/maintenance.md	Maintainer commands for validation, packaging, host cache checks, and local development.
docs/architecture.md	Framework layers and runtime entrypoints.
docs/adding-skills.md	How to add or adapt skills.
docs/platform-support.md	Host adapter expectations and support notes.

Development Contract

• Keep canonical skills in skills/<skill-name>/SKILL.md.
• Keep callable agent sources in agents/.
• Keep orchestration protocols in workflows/.
• Keep host-specific metadata in adapter directories only.
• Keep provider and source governance in registry/.
• Regenerate integrations/ from agents/; generated files are derived outputs.
• Keep private configuration out of the package.

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