Creator workflow — ship a skill and earn.

The LoopSkill creator program lets any developer publish skills to the marketplace and earn a share of subscription revenue. This guide walks through the full journey — from signup to first payout — in the order you'll actually do it.

Step 1 — Signup and get your API key

Create an account at recipes.wisechef.ai. Then, from the dashboard → API Keys → New key, create a rec_ key.

export RECIPES_API_KEY=rec_xx...   # add to your ~/.bashrc or .zshrc

You don't need a paid subscription to submit a skill — submissions are open to all API key holders. A subscription is only required to install cook-tier skills.

Step 2 — Design your skill

A good skill has these properties:

• Solves a real recurring problem. "I do this manually every week" is the gold standard. One-off tasks don't make great skills.
• Uses env vars for credentials. Never hardcode API keys, paths, or usernames. The skill must work out-of-the-box on any machine.
• Has a clear, testable output. A report file, a Git commit, a formatted JSON response — something the agent can verify succeeded.
• Is architecture-aware. If your skill needs CUDA, declare it in the frontmatter's hardware_requirements block. The install runtime will check compatibility before downloading.

Step 3 — Scaffold the skill locally

Use the recipes_recipify MCP tool to turn a description into a SKILL.md draft:

// Ask your agent:
"Recipify a skill that audits a website's Core Web Vitals and outputs a PDF report"

// Or scaffold manually:
mkdir -p ~/skills/my-skill/{references,templates}
touch ~/skills/my-skill/SKILL.md
touch ~/skills/my-skill/skill.toml

The minimal SKILL.md frontmatter:

---
name: My Skill
description: >
  One-sentence description of what this skill does and who it's for.
  Under 200 chars. This is what users see in search results.
version: 1.0.0
tier: cook        # free | cook | operator
license: MIT      # SPDX identifier
category: code    # marketing | code | web-scraping | ops | sales | sim-robotics
---

# My Skill

## What this does
...

## Prerequisites
- env: MY_API_KEY

## Steps
1. ...
2. ...

## Output
...

Step 4 — Test locally

Install your skill locally to test it before submitting. The easiest path is to point your agent directly at the local SKILL.md:

# Copy to your agent's skills dir and test
cp -r ~/skills/my-skill ~/.claude/skills/my-skill

# Then ask your agent to run it
"Run my-skill on example.com"

Alternatively, use the CLI to do a local install from the directory:

recipes install my-skill --install-dir ~/.claude/skills

Iterate until the skill produces the expected output on a clean run. Check that it fails gracefully when env vars are missing (your skill should print a helpful error, not crash silently).

Step 5 — Submit via recipes_publish_request

Once your skill works locally, submit it from your agent conversation:

// Ask your agent:
"Submit my skill at ~/skills/my-skill/ to the LoopSkill marketplace"

// The agent reads the directory and calls:
recipes_publish_request({
  slug: "my-skill",
  content: "<SKILL.md contents>",
  version: "1.0.0",
  description: "One-sentence description",
  tier: "cook",
  license: "MIT"
})

The tool runs the quality gate synchronously. If it passes, you get back:

{
  "request_id": "uuid...",
  "slug": "my-skill",
  "status": "pending_review",
  "issue_url": "https://github.com/wisechef-ai/recipes-api/issues/NNN",
  "sha256": "abc123...",
  "warnings": []
}

The issue_url is your tracking link. Bookmark it — this is where the review happens.

Step 6 — Respond to review feedback

A WiseChef team member will review your skill on GitHub. Common feedback:

• Quality gate warnings — fix the flagged patterns and resubmit. Use the same slug; the system links the new submission to the open issue.
• "Needs a clearer output contract" — add an ## Output section to your SKILL.md describing exactly what the skill produces.
• "Tier mismatch" — the reviewer may suggest a different tier. Update your skill.toml and resubmit.
• "Please add telemetry hooks" — add the standard telemetry block to your skill's final step. See the telemetry docs.

To resubmit after changes, call recipes_publish_request again with the same slug. Rate limit is 1 submission per 24h per (user, slug) — if you need to resubmit faster, comment on the GitHub issue.

Step 7 — Approved and published

When approved, the reviewer closes the issue with approved-publish. Within minutes:

1. The tarball is built from your approved submission
2. Signed with the WiseChef ed25519 key
3. Uploaded to the CDN
4. Your skill appears in search results
5. You receive a confirmation email with the catalog URL

Step 8 — Maintaining your skill

Published skills need maintenance. The auto-improve incident network clusters anonymous failure telemetry across the fleet. Skills with high failure rates appear in your creator dashboard with incident details.

To ship a patch: increment the version in your SKILL.md frontmatter and call recipes_publish_request again. Patch versions (e.g. 1.0.1 → 1.0.2) go through an expedited review. Major versions (1.x → 2.0.0) get a full review.

Users on a Pro subscription who have your skill in a bundle will see the update the next time they run recipes_sync or recipes sync.

Creator earnings

Once published, your skill earns a share of subscription revenue attributed to its install and use events. Attribution is based on the telemetry event stream — every skill_install and skill_use event from your skill's slug is counted.

See the Creator program page for current revenue share percentages, payout minimums, and the Stripe Connect payout setup.

Related docs

• Publishing reference — quality gate details, rate limits
• API reference — endpoint docs
• Security model — review pipeline, ed25519 signing
• Creator program — earnings, referrals, payout