Skill Creator: Build SKILL.md Agent Skills with Orthogonal in 2026
Orthogonal's Skill Creator teaches AI agents to design, structure, and publish SKILL.md packages in 5 steps. Free CLI, open source, works with Claude Code.
Orthogonal's Skill Creator is a SKILL.md package that teaches AI agents how to build, structure, and publish other SKILL.md skills using the Orthogonal CLI. It covers a 5-step creation process, a degrees-of-freedom framework for matching format to task fragility, and a token-efficiency philosophy. The creation workflow itself is free; the orth CLI is included with an Orthogonal account. Works with Claude Code and Cursor.
Orthogonal's Skill Creator is a meta-skill (a SKILL.md file) that teaches AI agents how to design, structure, and publish other SKILL.md agent skills. It covers a 5-step process: understand requirements, plan reusable scripts and references, initialize with orth skills init, implement the SKILL.md, and submit via orth skills submit. Compatible with Claude Code, Cursor, and any orth CLI user. The creation workflow is free; published skills call paid APIs.
Maker: Orthogonal · Protocol: CUSTOM · Auth: api key
Compatible agents: Claude Code, Cursor, Any orth CLI-compatible agent, OpenClaw, Any MCP-compatible client
Required runtime: Node >= 18 (for orth CLI), Orthogonal account
About Skill Creator
Orthogonal's Skill Creator is a meta-skill: a SKILL.md file that teaches AI agents how to design, structure, package, and publish other SKILL.md skills. Rather than a traditional API connector or data integration, it is a workflow guide for agents building new agent capabilities. The skill covers the full skill creation lifecycle: gathering concrete usage examples, planning reusable scripts, references, and assets, initializing a skill directory with `orth skills init`, writing the SKILL.md with correct frontmatter and progressive disclosure structure, and submitting the skill to the Orthogonal catalog via `orth skills submit`. It is part of the orthogonal-sh/skills open-source repository maintained by Orthogonal, a YC W26-backed company. A SKILL.md file is a markdown document with a YAML frontmatter block (name, description) followed by structured instructions the agent reads when the skill is triggered. The description field serves as the trigger mechanism: the agent's runtime matches user requests against skill descriptions to decide which skill to load. This means all 'when to use' guidance must be in the frontmatter description, not the body. The Skill Creator covers this design constraint explicitly and explains how to match specificity to task fragility: high-freedom text instructions for tasks with many valid approaches, pseudocode or parameterized scripts for preferred patterns, and specific scripts for fragile or error-prone operations. The skill enforces a token-efficient philosophy: every paragraph must justify its context window cost, SKILL.md files should stay under 500 lines (splitting into references/ files for complex domains), and README, changelog, installation guide, and setup procedure documents are explicitly excluded from skill packages. Skills can include a scripts/ directory for deterministic executable code, a references/ directory for domain-specific docs loaded on demand, and an assets/ directory for templates or images used in output. The Orthogonal CLI provides three commands for the skill lifecycle: `orth skills init <name>` creates the directory scaffold, `orth skills submit <path>` publishes to the Orthogonal catalog for discovery by other agents, and `orth skills update <slug> <path>` updates a published skill. Orthogonal's pricing model is pay-per-call; the Skill Creator SKILL.md itself does not directly consume credits (it is instructions, not an API call), but CLI commands and any sub-APIs called by the skills it creates do. Orthogonal was accepted into YC W26, founded by Christian Pickett (ex-Coinbase payments, ex-Vercel billing) and Bera Sogut (ex-Google reCAPTCHA/Maps, ex-Amazon Robotics). The orthogonal-sh/skills repository was last updated in June 2026 and contains 90+ skills covering communication, developer tools, data research, and productivity integrations.
Key Features
- SKILL.md Structure Guide: Defines the exact SKILL.md format: YAML frontmatter with name (lowercase, hyphens, under 64 chars) and description (trigger mechanism), followed by instructions with optional scripts/, references/, and assets/ directories.
- Degrees-of-Freedom Framework: Provides a 3-level specificity model: high-freedom text instructions for open-ended tasks, medium-freedom pseudocode for preferred patterns, and low-freedom specific scripts for fragile operations, so skill authors match format to task fragility.
- 5-Step Creation Process: Walks agents through understand (gather usage examples), plan (identify reusable scripts and references), initialize (`orth skills init <name>`), implement (write SKILL.md and add resources), and submit (`orth skills submit <path>`) in sequence.
- Progressive Disclosure Pattern: Teaches the high-level guide with references/ pattern: keep SKILL.md under 500 lines by splitting complex domain docs into separate reference files that the agent loads on demand, reducing context window usage for skills not fully invoked.
- CLI Publishing Commands: Covers `orth skills init <name>` (scaffold), `orth skills submit <path>` (publish to Orthogonal catalog), and `orth skills update <slug> <path>` (update existing), with optional --path flag to target a specific skills directory.
- Token Efficiency Rules: Enforces a 'concise is key' philosophy: exclude README, changelog, setup, and testing documentation from skill packages; challenge every paragraph for context window cost; prefer examples over verbose explanations; and cap SKILL.md at 500 lines.
How to Invoke
Agent reads SKILL.md instructions and follows the 5-step creation process: understand, plan, initialize with `orth skills init`, implement, and submit with `orth skills submit`
Pricing
The Skill Creator SKILL.md is free and open source. The orth CLI commands (init, submit, update) are included with an Orthogonal account. New accounts get $10 in free credits. Credits apply to APIs called by the skills you create, not to the creation workflow itself.
Strengths
- Teaches agents a documented, repeatable framework for building any new skill, reducing inconsistency in how agent capabilities are packaged across a team or project.
- The progressive disclosure pattern (SKILL.md under 500 lines, with references/ for complex domains) keeps individual skill context overhead small for agents that need to load multiple skills simultaneously.
- Open source SKILL.md in the orthogonal-sh/skills repo serves as a reference for skill authors even if they do not use the Orthogonal CLI or catalog.
Weaknesses
- The skill teaches a workflow through SKILL.md instructions, not executable code; if the agent misreads or partially follows the instructions, the resulting skill may have incorrect frontmatter or missing structure.
- Published skills in the Orthogonal catalog are only discoverable by other orth CLI or Orthogonal MCP users; skills cannot be published to other skill directories (Claude skills marketplace, Cursor rules index) through this workflow.
- The 500-line SKILL.md limit and token-efficiency rules require skill authors to make editorial decisions about what to include, which adds a judgment overhead not present in simpler API wrapper skills.
Frequently Asked Questions
What is Skill Creator (via Orthogonal) and what does it do?
Orthogonal's Skill Creator is a meta-skill: a SKILL.md file that teaches AI agents how to design, structure, and publish other SKILL.md skills for agent capability extension. It is part of the orthogonal-sh/skills open-source repository maintained by Orthogonal, a YC W26-backed company, alongside 90+ other integration and automation skills. The skill defines the canonical SKILL.md format: a YAML frontmatter block with name (lowercase, hyphens, under 64 chars) and description (the trigger mechanism), followed by structured instructions and optional scripts/, references/, and assets/ subdirectories. It covers a 5-step creation lifecycle: understand requirements, plan reusable resources, initialize with `orth skills init <name>`, implement the SKILL.md, and publish with `orth skills submit <path>`. A degrees-of-freedom framework guides format selection: high-freedom text for open-ended tasks, medium-freedom pseudocode for preferred patterns, and low-freedom scripts for fragile operations. The skill also teaches a token-efficiency philosophy: keep SKILL.md under 500 lines, split complex docs into references/ files, and exclude setup guides and changelogs from skill packages.
How much does Skill Creator (via Orthogonal) cost in 2026?
The Skill Creator SKILL.md itself is free and open source in the orthogonal-sh/skills GitHub repository. The orth CLI commands used in the creation workflow (orth skills init, orth skills submit, orth skills update) are included with a free Orthogonal account and do not consume credits. New Orthogonal accounts receive $10 in free starting credits, which apply to API calls made by the skills you create (not to the creation workflow itself). Creating and publishing a new skill has no direct credit cost; running a skill that calls paid APIs (Exa, Scrape Creators, Fiber, etc.) is what depletes credits. There are no fees for publishing to the Orthogonal skills catalog; Orthogonal does not charge a marketplace listing fee. The only ongoing cost is credit consumption when the finished skills are executed by agents.
What are the main features of Skill Creator (via Orthogonal)?
The core feature is a documented 5-step skill creation process: understand (gather concrete usage examples), plan (identify scripts, references, and assets), initialize (run `orth skills init <name>`), implement (write SKILL.md and add resources), and submit (run `orth skills submit <path>`). The degrees-of-freedom framework is the key design decision tool: high freedom (text instructions) for tasks with many valid approaches, medium freedom (pseudocode) when a preferred pattern exists, and low freedom (specific scripts) for fragile or error-prone operations. The progressive disclosure pattern keeps skills under 500 lines by splitting complex domain docs into separate references/ markdown files that the agent loads on demand. The CLI covers `orth skills init` (scaffold), `orth skills submit` (publish), and `orth skills update <slug>` (version updates). Token-efficiency rules exclude README, changelog, setup, and test documentation from skill packages. The skills/ directory structure supports scripts/, references/, and assets/ as optional subdirectories alongside the required SKILL.md.
Is Skill Creator (via Orthogonal) free to use?
The Skill Creator SKILL.md is fully free: it is open source in the orthogonal-sh/skills GitHub repository and can be read and used as a reference without any account. The orth CLI commands for skill initialization and publication (orth skills init, orth skills submit, orth skills update) are free to use with an Orthogonal account and do not consume credits. New Orthogonal accounts receive $10 in free credits, but those credits apply to API calls from executing skills, not from creating them. Publishing skills to the Orthogonal catalog has no listing fee. The creation workflow itself has no cost beyond the time it takes the agent to follow the instructions. Any cost only comes into play when the skill you create invokes paid Orthogonal sub-APIs (like Exa, Scrape Creators, or Fiber) during execution.
What are the best alternatives to Skill Creator (via Orthogonal)?
Claude's built-in skill creation workflow (writing a SKILL.md manually based on the Anthropic documentation) is the most direct alternative, requiring no Orthogonal account and following a similar SKILL.md format. The Anthropic skills framework for Claude Code supports a similar SKILL.md format with frontmatter-based triggers and supports publishing to the Claude skills directory independently of Orthogonal. Cursor Rules (.cursorrules files) serve a similar purpose for Cursor-specific agent behavior customization, but do not use SKILL.md format and are not interoperable with other agents. OpenClaw's skill format is another option for OpenClaw users, following a compatible but distinct structure. The key advantage of Orthogonal's Skill Creator over writing a SKILL.md from scratch is the documented degrees-of-freedom framework, token-efficiency guidelines, and direct CLI integration for publishing to the Orthogonal catalog. If you only use Claude Code and do not need Orthogonal CLI integration, writing directly from Anthropic's skill documentation is simpler.
Who is Skill Creator (via Orthogonal) best for?
This skill is best for developers building reusable agent capabilities that they want to package in a documented, shareable format compatible with the Orthogonal CLI and catalog. Teams building internal agent workflows in Claude Code or Cursor who want consistent skill structure across their codebase will benefit from the degrees-of-freedom framework and token-efficiency rules. Developers contributing to the Orthogonal open-source skills catalog who need to follow the canonical submission format (frontmatter structure, directory layout, 500-line limit) are a clear fit. It is not the right tool for users who only need a one-off prompt or system message; the overhead of init/implement/submit is only worthwhile for reusable capabilities. Non-technical users who want to customize agent behavior without writing structured markdown files are better served by visual prompt builders. Enterprise teams who need skill versioning, access control, and audit logs should evaluate whether Orthogonal's catalog meets their governance requirements.
How do you get started with Skill Creator (via Orthogonal)?
Sign up at orthogonal.com for a free account (no credits needed just to create skills). Install the orth CLI with `npm install -g @orth/cli` in a terminal with Node 18 or later, then log in with `orth login`. Install the Skill Creator skill with `orth skills add orthogonal-skill-creator`, placing the SKILL.md instructions in your agent's skills directory. Open Claude Code or Cursor and ask your agent to create a new skill, providing a concrete description of what the skill should do and 2-3 example use cases. The agent will follow the 5-step Skill Creator workflow: gathering your requirements, choosing the appropriate degrees-of-freedom level, running `orth skills init <your-skill-name>`, writing the SKILL.md with correct frontmatter and instructions, and submitting with `orth skills submit ./your-skill-name`. Review the generated SKILL.md before submission to verify the frontmatter description accurately captures the trigger condition, and test the skill on at least 1-2 real tasks before sharing.
How does Skill Creator compare to writing a SKILL.md manually in 2026?
Writing a SKILL.md manually from Anthropic's or Orthogonal's documentation is faster for simple, one-off skills but produces less consistent structure without the Skill Creator's guidance. The Skill Creator adds value through 3 specific decisions that are easy to get wrong manually: choosing the correct degrees-of-freedom level (text vs. pseudocode vs. script), deciding what to split into references/ files to stay under the 500-line limit, and writing a frontmatter description that triggers correctly without including body content the agent won't see at trigger time. Manual SKILL.md writing skips the `orth skills init` scaffolding step, which means the scripts/, references/, and assets/ directories may be missed or inconsistently named. The Skill Creator also enforces the 'no README, no changelog, no setup docs' rule explicitly, which manual authors often violate by adding explanatory content that bloats the skill package. For experienced skill authors who have internalized the format, manual writing is equally effective; for new contributors to the Orthogonal catalog, the Skill Creator reduces first-attempt errors.