skills

Imported from mulino-coreano/mulino-coreano-erp (agents/skills/AGENTS.md). Install upstream with npx skills add mulino-coreano/mulino-coreano-erp --skill skills. Copyright stays with the author.

agents/skills/ — agent role skills

One directory per agent role, each holding a SKILL.md. These are authored here as the SSOT and symlinked into both discovery directories — .agents/skills/ for Codex and .claude/skills/ for Claude Code — so sessions auto-discover them. The contracts below are already in effect even though the mulino CLI is not built yet — treat mulino references as the command surface these roles will be bound to, not as runnable today.

Required sections in every SKILL.md

1. Mission — the role's one-paragraph purpose
2. Allowed commands — which mulino subcommands this role may use (see ../cli/AGENTS.md for the contract)
3. Governance expectations — which of its writes return PENDING_APPROVAL and what to do then (e.g. Procurement: PO creation always pends MANAGER approval — never treat it as failure)
4. Korea localization invariants it guards — from the root AGENTS.md
5. Hand-off triggers — situations where the role reports back to the orchestrator for another role to take over

The orchestrator skill additionally defines the dispatch table: which role subagent handles which situation, and how results flow back.

Rules

• Role duties come from the agent intervention summary in docs/02_flow.md (SSOT) — don't invent new duties in a skill without updating the flow doc.
• Adding an agent = new folder + SKILL.md + a symlink into .agents/skills/ and another into .claude/skills/. No code.
• Skills contain role knowledge only — no credentials, no endpoint URLs (that's MULINO_API_URL), no SQL.