Agent Skills

Genfeed publishes a catalog of product skills: portable knowledge packages that teach an AI agent how to do content work. A skill is a folder with a SKILL.md file. The agent loads it, reads the instructions inside, and applies them when the task matches.

Skills carry knowledge, not credentials or code. They work in any agent that supports the SKILL.md convention, with or without a Genfeed account.

What a product skill is

A product skill encodes a repeatable content discipline — writing an X thread, scoring a draft against an SEO rubric, planning an editorial calendar, warming up a new social account. The skill supplies the framework, the rules, and the output format so the agent stops improvising.

Three things in the Genfeed source tree share the word “skill”. Only the first is something you install:

Thing	Where it lives	What it is
Product skills	`skills/` in the monorepo, published as `genfeedai/skills`	Content knowledge packages you install into your agent
Repository dev skills	`.agents/skills/`	Internal skills that help agents build, test, and ship Genfeed itself — not part of the product
Skill executor	`apps/server/api/src/services/skill-executor/*`	A backend service that runs deterministic generation handlers inside Genfeed

The catalog lives in skills/README.md, generated from each skill’s frontmatter — that file is the current list of what ships.

Install

Skills install with the skills CLI, which resolves any public GitHub repository that follows the SKILL.md convention.

Install the catalog


bunx skills add genfeedai/skills

Install a single skill by appending its slug:


bunx skills add genfeedai/skills/x-content-creator

Pick your agents

The CLI prompts for the agents to install into. Pass -a to skip the prompt:


bunx skills add genfeedai/skills -a claude-code -a codex

Claude Code, Codex, Cursor, Cline, OpenCode, and other spec-compliant agents are supported targets.

Choose a scope

By default skills install into the current project, under the agent’s own skills directory. Use -g to install them once for every project on the machine:


bunx skills add genfeedai/skills -g

Verify


bunx skills list

Then ask the agent something the skill covers — “write a tweet about our launch” — and it should pick up the matching skill without further prompting.

Keep skills current with bunx skills update, and drop one with bunx skills remove <name>.

Skills are plain Markdown directories. Any agent without CLI support can use them by copying the skill folder into whatever path that agent reads, or by pasting the SKILL.md body into its system instructions.

Authoring a skill

A skill is one directory containing a SKILL.md with YAML frontmatter and a Markdown body:


---
name: content-strategist
description: Build comprehensive content strategies including audience definition, content pillars, platform selection, and posting cadence. Triggers on "content strategy", "editorial calendar", "what should I post".
license: MIT
metadata:
  author: genfeedai
  version: '1.0.0'
---
 
# Content Strategist
 
You are an expert content marketing strategist and editorial planner...

Field	Required	Purpose
`name`	yes	Directory-matching identifier, kebab-case
`description`	yes	What the skill does and the phrases that should trigger it
`license`	no	SPDX identifier; Genfeed’s catalog is MIT
`metadata.author`	no	Publisher handle
`metadata.version`	no	Semver string, bumped on behavior changes

The description is the whole routing mechanism: the agent decides whether to load a skill from that line alone. Name the concrete phrases a user would say.

The body is the instruction set. Write it as directions to the agent — role, principles, a numbered process, output format, worked examples. Genfeed’s skills are long on process and short on prose.

Two optional files sit alongside SKILL.md:

README.md — human-facing summary and the install command for that one skill.
metadata.json — machine-readable triggers, tags, and outputs used for catalog and discovery surfaces.

Scaffold a new one with bunx skills init <name>.

Product skills belong in the monorepo’s root skills/ directory. Skills that help agents work on the Genfeed codebase itself belong in .agents/skills/. Mixing the two ships internal build guidance as product content.

Composing with the MCP server

Skills and the MCP server solve different halves of the same problem.

A skill is knowledge. It tells the agent what good output looks like and how to get there. It needs no account and no network access.
The MCP server is capability. It gives the agent authenticated tools to generate images and video, run workflows, publish posts, and read analytics from your Genfeed workspace.

Used together, the agent plans with the skill and executes with the tools: the x-content-creator skill shapes the thread, then the MCP create_post tool schedules it. The openclaw-integration skill exists specifically to teach an agent how to drive the Genfeed MCP surface.

Connect the server once and every installed skill can reach it:


claude mcp add --transport http genfeed --scope user https://mcp.genfeed.ai/mcp --header "Authorization: Bearer $GENFEED_API_KEY"

Agents that only need discovery — no key, no tool calls — can read genfeed.ai/.well-known/agent.json, which advertises the agent’s headline capabilities and its MCP and REST entry points.

Next steps

MCP Server — endpoints, authentication, and client setup
How to prompt AI content: the CLEAR framework — the framework skills build on
Skill catalog — every published product skill