Skills

MemoryMarket ships two Claude Code / Codex / Cursor skills inside the memorymarket npm package. Skills let your AI agent drive the CLI end-to-end — you only confirm decisions, the agent runs every command. Two skills cover the full lifecycle: /mm-publish (creator) and /mm-add (buyer). Pack extraction runs inline inside /mm-publish Phase 3; wiki integration runs inline inside /mm-add Phase 4 — there are no separate skills for those.

Install the skills

The skills ship in the npm package under skills/. Install the CLI, then ask your agent to copy them into its skills directory:

$ npm install -g memorymarket
# then in your agent:
Install the MemoryMarket skills from the npm package.

The agent locates the skills inside the installed package, copies them to ~/.claude/skills/ (Claude Code) or the equivalent path for your agent, and confirms when both are active.

The two skills

/mm-publish

Creator path. Drives the publish flow — auth gate, signing key setup, memory source discovery, pack assembly, and publish. The user creates the draft on the web at /sell/new first, copies the pack-id from the URL, and pastes it into /mm-publish --pack-id <uuid>. The skill never creates drafts; that always happens on the web.

/mm-add

Buyer / installer path. Wraps mm install and an optional inline wiki-integration step into one conversational flow. Asks for the pack-ref and scope, runs the install, patches the host agent rules file with a managed block, and optionally weaves the pack into your knowledge wiki (Phase 4 runs the wiki integration inline — no separate skill needed).

How the skills relate to the raw CLI

The skills are conversational wrappers around the same mm CLI commands you can run by hand. Anything a skill does, you can do with a sequence of CLI calls — the skill exists so you do not have to memorize the order or remember when to run mm whoami vs mm login vs mm keygen. See MM-AMP Integration for the raw CLI reference.

Source of truth

Each skill page below mirrors the SKILL.md file shipped inside the npm package. If the docs and the SKILL.md disagree, the SKILL.md wins — that is what your agent actually reads at invocation time.