Agent skills glossary
Short definitions of the terms used across Skill Store 33. Terms from outside standards follow their official documentation.
Skills
Agent skill
A folder of instructions, and optionally scripts and reference files, that an AI agent loads when a task calls for it. The format is defined by the Agent Skills specification (agentskills.io) and is supported by Claude Code, Codex, Cursor and other agents.
SKILL.md
The one required file in a skill. It starts with YAML frontmatter, followed by Markdown instructions the agent follows once the skill is active.
Frontmatter
The block of YAML between --- lines at the top of SKILL.md. The specification requires name (up to 64 characters: lowercase letters, numbers and hyphens, matching the folder name) and description. Optional fields include license, compatibility, metadata and allowed-tools.
Description
The frontmatter field that says what a skill does and when to use it, in up to 1,024 characters. Agents read it to decide whether to load the skill, and Skill Store 33 search matches tasks against it.
Progressive disclosure
How agents load skills in stages to save context. At startup an agent loads only each skill's name and description. It loads the full SKILL.md when it activates the skill, and loads other files in the folder only when needed.
Skill folder
The directory that holds a skill: SKILL.md plus optional scripts/, references/ and assets/ folders. Its name should match the skill's name field.
.agents/skills
A folder where some agents look for skills. Codex scans .agents/skills in a repository, from the working directory up to the repository root, and in your home directory. Claude Code uses .claude/skills instead.
Plugin
In Claude Code, a package that bundles skills with other extensions such as subagents, hooks and MCP servers. Skills from a plugin are namespaced as plugin-name:skill-name.
Subagent
A separate agent instance that the main agent hands a task to. It works in its own context window and does not see the main conversation.
Connecting agents
MCP
The Model Context Protocol, an open standard for connecting AI applications to external tools and data. A client inside the agent talks to servers that offer tools, resources and prompts.
MCP server
A program or web service that offers capabilities to agents over MCP. Skill Store 33 runs one at https://mcp.ss33.ai.
Streamable HTTP
The MCP transport for remote servers. The client sends JSON-RPC messages to a single HTTP endpoint, and the server answers with JSON or a stream of server-sent events.
Tool call
A request from the model to run a named tool with arguments, such as search_skills with a query. The result goes back into the model's context.
SEP-2640
The MCP Skills Extension proposal, now accepted. It lets an MCP server offer skills as MCP resources under the extension id io.modelcontextprotocol/skills.
AGENTS.md
An open format: a Markdown file in a repository that gives coding agents project context and instructions. It is stewarded by the Agentic AI Foundation under the Linux Foundation. Unlike a skill, it is read for every task in that project.
CLAUDE.md
Claude Code's project memory file. Claude Code loads it into context at the start of a session, while a skill's body loads only when the skill is used.
Context window
The amount of text a model can consider at once, measured in tokens. Instructions, files, tool results and the conversation all share it.
Tokens
The units models read and write: whole words or pieces of words. Skill Store 33 shows each skill's approximate size in tokens, estimated as its length in characters divided by four.
In the store
Verification states
The trust label on each listing: Not yet verified (untested), Safety scanned (a safety scan only), Verified (the skill's own tests ran in a sandbox and passed) or Tests failing.
Safety scan
An automated safety check of a skill's text. A skill marked Safety scanned has passed one, but nobody has tested whether it works.
Licence
The terms under which a skill may be used and shared, taken from its license field or the repository's licence file. Skill Store 33 serves a skill's text only under a licence that allows redistribution, and otherwise links to the author's copy.
Copyleft
Licences such as the GPL or CC BY-SA, which require copies and derived works to carry the same licence. Skill Store 33 lists copyleft skills but does not redistribute their text.
Publisher
The account a skill is listed under. For skills imported from GitHub, it is the account or organisation that owns the repository.
Skill Store 33 key
A free key you create on your Skill Store 33 account page. Send it to the MCP server as an Authorization: Bearer <key> header to raise your search limit from 5 to 60 searches a minute.
Bittensor and SN33
Bittensor
An open network of independent subnets that produce digital commodities. Participants are paid in TAO in proportion to the value they contribute.
Subnet
An independent Bittensor network that produces one kind of output. Its creator defines the incentive mechanism that decides how work is scored.
SN33
Bittensor subnet 33, run by ReadyAI, the company behind Skill Store 33. It has a task for writing skills with test plans, and it is one of the store's sources, with 13 skills so far.
Miner
A participant in a Bittensor subnet that produces the subnet's output, such as a skill and its tests on SN33.
Validator
A participant in a Bittensor subnet that scores miners' work. The scores decide how rewards are shared.