Skills

Arcantry skills are self-contained Agent Skills packages. Each one owns a focused procedure, public metadata, scenarios, and any references or scripts it needs.

Three families

Family	Use it for
self-improvement	Capture repeated work, create and evaluate skills, maintain agent guidance, select task skills, and stage review findings.
repo-safely	Adopt a repository, capture incoming work, promote todo work into OpenSpec, reconcile sources, maintain release meaning, and verify proportionally.
content-safely	Protect audience and privacy, design terminal experiences, and write concrete product content without slop.

No family is a router skill. Every catalog entry performs one focused job.

Inspect and link one skill

arcantry skills list
arcantry skills inspect <name>
arcantry skills link <name> --scope user
arcantry skills link <name> --scope user --compat claude
arcantry skills doctor --scope user

Use user scope by default so the skill remains available across projects. Choose repository scope only for a shared repository requirement, and use a private package under .local/skills only as an explicitly selected fallback or local override.

Arcantry recommends the universal Agent Skills locations:

Surface	User scope	Repository scope
Universal Agent Skills	~/.agents/skills	<repo>/.agents/skills
Optional Claude compatibility	~/.claude/skills	<repo>/.claude/skills

Codex reads the universal surface directly. Claude Code compatibility is an additional alias to the same canonical package:

arcantry skills link <name> --scope repo --compat claude

Use --target <path> only for an advanced explicit Agent Skills directory. It cannot be combined with --scope or --compat.

Linking is idempotent when the target already points to the canonical package. Arcantry preflights the universal and compatibility destinations before writing. It does not overwrite an ordinary directory. --replace creates a backup first. Unlinking removes only exact links to the selected skill.

Keep repository skills private

A repository can keep a canonical skill under .local/skills/<name> and expose it through locally excluded links:

arcantry skills list --scope private
arcantry skills inspect <name> --scope private
arcantry skills link <name> --scope private
arcantry skills link <name> --scope private --compat claude

Private and public packages cannot reuse the same skill name. .agents and .claude aliases that resolve to one real package remain one skill, not duplicates.

Load the complete catalog

The repository exposes the same skills/ tree through optional package manifests:

claude --plugin-dir ./arcantry

Claude Code namespaces plugin skills as /arcantry:<name>. Codex can use .codex-plugin/plugin.json from the same repository. Both manifests carry the current Arcantry version, but neither is required for the universal .agents workflow.

Other Agent Skills workflows

The packages use the open Agent Skills directory format, so compatible independent installers can discover them from skills/*/SKILL.md:

gh skill install MrMaxie/arcantry <name> --agent codex --scope user
npx skills add MrMaxie/arcantry --skill <name> -a codex -g

See the GitHub CLI skill installer and the open skills CLI for their current scope and host rules. They may choose host-specific destinations. Manual copying or symbolic linking also works.

These workflows are alternatives to the Arcantry linker, not runtime dependencies.

Tool and write boundaries

A skill may declare a connector or command-line dependency. That declaration does not authorize creating issues, posting replies, publishing content, or changing another system. External writes still require authority for the exact target and action.

Generated reference

Per-skill pages are generated from canonical packages so descriptions, scenarios, family placement, and dependencies remain aligned.

Browse the complete skill catalog