skills-source.md

title	Skills — Source of Truth
description	Skills are the callable surface that any operator (FlowPilot, OpenClaw, Claude Desktop, custom MCP client) uses to act on FlowWink.
category	reference

Skills — Source of Truth

Skills are the callable surface that any operator (FlowPilot, OpenClaw, Claude Desktop, custom MCP client) uses to act on FlowWink.

Where they're defined

Each module declares its skills in skillSeeds on its defineModule() manifest:

src/lib/modules/<module>-module.ts → skillSeeds: SkillSeed[]

When a module is enabled, the bootstrap seeds those rows into public.agent_skills with mcp_exposed=true (unless the module marks them operator-internal). The database table is the runtime registry — hot-reloadable, RLS-protected, and the single thing agent-execute and the MCP server read.

How an operator discovers them

Surface	How
In-app admin	/admin/developer → Skills tab — full catalog with descriptions, schemas, gating
MCP (external agents)	`GET /functions/v1/mcp-server/rest/tools[?groups=marketing
Per-module reference	Every page in ../modules/ lists that module's skills with handler + scope
Live count	select count(*) from agent_skills where enabled and mcp_exposed

Adding a skill

1. Add a SkillSeed to the module's skillSeeds array.
2. Pass Agent Contract Integrity — every NOT NULL DB column the action writes must be in the JSON schema (bun run lint:skills).
3. Write description with explicit Use when: and NOT for: markers (Law 2 — skills are self-describing).
4. Disable & re-enable the module, or run the "Sync Skills" action in /admin/developer.

Built-in tools

agent-reason provides a small set of hardcoded operator-internal tools (memory ops, objective planning, reflection, soul, skill-pack management, delegation). These are FlowPilot-only and are not stored in agent_skills — they are the operator's own brain primitives, not platform capabilities.

---

See also: Module API · MCP as Platform · Agent Contract Integrity