A shared library and route map
Every AI knows how—and who can do it.
Virtual Skills let the fleet reuse complete instructions. Virtual Capabilities reveal which runtime owns the plugin or extension that can execute a task. Captain Memo keeps both catalogs local, searchable and synchronized.
Skills and plugins
The files, extensions and runtimes each CLI already owns.
Captain Memo
Full skill instructions plus sanitized, non-executable capability descriptors in SQLite.
Any connected AI
Load the right method or delegate work to the runtime that owns the tool.
How it sails
Discover. Choose. Load.
The catalog stays light until a skill is genuinely useful. Assistants see small descriptors first and retrieve the full instruction body only after choosing one.
Discover automatically
If the setting is missing, Captain Memo scans the standard skill roots for Claude Code, Codex, Gemini, Cursor, opencode, Vibe and Kimi. An explicitly empty setting is the opt-out.
Browse or recommend
list_skills shows the catalog. recommend_skills ranks the few skills relevant to the task; Captain Fleet can search sibling captains too.
Load deliberately
load_skill returns the complete, unabridged instructions. They are advisory: system, user, repository and native skill instructions always remain authoritative.
See what the crew knows
List the virtual skills.
Humans get a terminal command; connected assistants get the matching MCP tool.
$ captain-memo skill list release-checklist [claude-code] Publish and verify a release skill:release-checklist:… code-review [codex] Review changes for correctness skill:code-review:…
list_skills({ source_agent: "codex", limit: 50 }) // then, when one fits: load_skill({ doc_id: "skill:code-review:…" })
Virtual Capabilities
Know which AI can actually do the work.
Skills are portable advice. Plugins and extensions are runtime-specific executors. Captain Memo turns their safe public surface into a fleet-wide route map: an assistant can recommend Nano Banana for image generation, see that Gemini owns it, and delegate there.
Discover automatically
Captain Memo watches supported Claude, Codex, Gemini and Agy plugin manifests by default. Updates and removals follow the installed source.
Publish only the safe surface
The catalog keeps names, descriptions, versions, operation names and interfaces. It never copies environment values, credentials, raw manifests, command arguments or executable configuration.
Delegate to the owner
recommend_capabilities returns the owning runtime—and, in Captain Fleet, the owning captain. The caller routes the job there instead of trying to import or execute foreign plugin code.
See what the fleet can do
List and recommend capabilities.
The same sanitized catalog is available to humans and connected assistants.
$ captain-memo capability list nanobanana [gemini] Generate and edit images operations: generate, edit, diagram execution: delegate to gemini
recommend_capabilities({ task: "generate an image" }) // route, don't import execution: runtime: "gemini" mode: "delegate"
What “virtual” means
A mirror and a map—not a mystery copy.
The original SKILL.md or plugin manifest remains the source of truth. Captain Memo makes the useful surface searchable, portable and safe to share without pretending every CLI uses the same native format or can execute the same plugins.
It updates live
Change an installed skill or capability and Captain Memo refreshes the matching database row. Delete the source and its virtual entry disappears.
It travels with backups
Both catalogs live in Captain Memo’s local database, so the existing backup and restore path carries them with the rest of the corpus.
Its origin stays visible
Every entry records its source AI and provenance. Capabilities also keep an explicit delegation route to the runtime that can execute them.
It does not seize command
Imported skill instructions remain advisory, and capability descriptors remain non-executable. Neither overrides user or repository policy.