A folder containing a SKILL.md file: Markdown instructions with a short block of metadata at the top saying what the skill is for and when an agent should reach for it, plus any scripts, templates or reference files it needs. It is the Agent Skills specification, which Anthropic published in December 2025 and released as an open standard. An agent loads the description at startup, decides from it whether the skill is relevant, and reads the rest only when it is.

Around forty agent products read the format, including Claude and Claude Code, ChatGPT and Codex, GitHub Copilot, VS Code, Cursor, Gemini CLI, JetBrains Junie, Amp, Goose, Mistral Vibe, Kiro, Roo Code, Tabnine, Databricks and Snowflake. Tools differ a little in where they expect skills to live, so we hand yours over with the exact path for each one your team runs.

Yes. The skill is yours, in a published format, in your repository, to edit and version like any other file you own. Third-party source material stays under its own licence: we cite and link it rather than re-publishing it as ours.

Ten working days from the point we accept the brief. We read every brief before starting, and if something needs clarifying we email you rather than guess.

Yes, and the skill is better for it. Handbooks, runbooks, style guides, process notes, past deliverables, proposals, specifications, review comments: anything that describes how the work is actually done where you are. This is usually the difference between a skill that knows your field and one that knows your business.

Send a link and we fetch and transcribe it: internal talks, recorded workshops, training sessions, webinars, conference presentations, podcast episodes, screen recordings with narration. Video does not belong in a browser upload, so links are the route for anything long. Slide decks, PDFs and spreadsheets go through the upload box with the rest.

PDF, Word, ePub, RTF, plain text, Markdown and CSV through the form, up to ten files at 25MB each. Links for recordings, video and anything on the web. If you have something larger or in an unusual format, ask before ordering and we will tell you whether we can read it.

It is used to build your skill and nothing else. Documents you upload are parsed in an isolated environment with no network access, because material we did not choose is material we do not trust by default. Please do not send anything you are not free to share with us. If you have a confidentiality requirement, raise it before ordering so we can agree how the material reaches us and what happens to it afterwards.

That is the normal case and it changes nothing about the price. Our researchers read your topic area in detail either way. Your material makes the skill more specific to you; it is not a prerequisite.

For something small, do. Just know what it produces: a model writing a skill writes from what it already knows, which is the one thing the agent did not need telling. The specification's own authoring guidance names this as the common failure, producing "vague, generic procedures" instead of the specific patterns and conventions that make a skill worth loading. The value is the part the model does not have, and that comes from your business, from the practitioners in your field, and from testing whether it changed the answer.

You set the test questions and we run each one twice, with the skill loaded and without it, and hand you both answers side by side. If a question came out the same either way, you will see that. It is the only way to tell a skill that works from a skill that reads well.

We tell you before delivery rather than padding the skill out. A thin subject makes a thin skill, and every claim in one is checked back against the source it came from, so what you get is what the material actually supports.

Then the assessment says so, and you have a score you can show someone the next time somebody asks whether it still works. We would rather tell you that than invoice you for a rewrite you did not need.

If the gap is the whole setup rather than one skill, that is a framework build: how your agents are wired, what they read, where they hand off. Those are scoped in conversation before any price is agreed. See what we have built.