Discovery & Validation

Most of these commands accept --json for machine-readable output.

care catalog

List installed AgentSkills, MCP servers, tools, and capability memory cards. Point it at the directories/files to enumerate.

care catalog --skills ./skills --mcp-config ./mcp_servers.toml --tools ./tools
care catalog --kind tool --json

Flag	Purpose
--skills DIR	Directory of <name>/SKILL.md folders (repeatable).
--mcp-config PATH	Path to an mcp_servers.toml.
--tools DIR	Directory of @carl_tool Python files.
--kind agent_skill|mcp_server|tool|memory_card	Filter to one entry kind.
--json	JSON instead of the grouped text listing.

care validate <file>

Parse and preflight a chain JSON file (bare-chain or wrapper form) — catch problems before running.

care validate chain.json
care validate chain.json --json

Flag	Purpose
--json	Emit the PreflightResult as JSON.

care import <pattern>...

Batch-validate chain JSON files and import them into Memory. Accepts paths or globs (**/*.json for recursive walks). It is the inverse of care export.

care import "chains/**/*.json"               # dry-run validate (default)
care import "chains/**/*.json" --apply        # actually save to Memory

Dry-run by default

care import only previews by default — it validates every match and reports what would be imported without touching Memory. Add --apply to actually save the valid chains. Exit code is 0 when every match validates, 1 otherwise.

Flag	Default	Purpose
--apply	off (dry-run)	Actually save chains to Memory.
--channel NAME	latest	Default channel for entries that don’t specify one.