# How skills work > How a skill is served, the Skills page and a skill's panel, the built-in skills, your profile, and the three tags memex reads. Page: https://docs.memex.tools/how-skills-work A skill is a note about how you want a job done, and memex advertises it to every connected assistant without ever applying it. Connected assistants are handed the list of skill titles and descriptions; when a job matches one, the assistant asks for that note and follows it. Nothing runs because it exists; it runs because the work matched it. ## Making one Tag a note `skill` and it is served once it is verified; a note your assistant proposes waits in the inbox first, like any note. Or press **New skill** on the Skills page, which opens the editor with the tag already set and a short scaffold (when to use it, steps, sources) for you to fill in. A `SKILL.md` from another tool comes in the same way as any note: hand it to your assistant, which proposes it as a skill. ## The Skills page **Skills**, in the sidebar, has three sections: **Your enabled skills**, **Available skills**, then **Built into Memex**. Your enabled skills are ready to use. Turning one off moves it into Available skills, alongside skills waiting for review. Turning it back on moves it into Your enabled skills. Failed saves show an error and switches return to the saved state. System skills appear at the bottom as compact cards with only their name, short description, and estimated token size. They cannot be edited or limited to particular connections, and only `memex-writing` can be switched off, in **Settings › Personalization**; while it is off, its card says so and links there. The summary counts enabled skills and shows their combined **Full instruction size**. The bar shows each skill's share of that total: muted built-ins first, then your enabled skills. Colors match the cards below. Hover, tap, or focus a segment to see its name, estimated tokens, and share of the total. Built-ins are always available and managed by Memex; your skills are yours to edit or turn off. Disabled, pending, and uninstalled skills are excluded. Turning a skill on or off updates the count, total, and bar. Each estimate is the instruction text's byte length divided by four, rounded up. Actual model token counts vary; custom system instructions may differ by connection. This is the size of the enabled library, not a context limit or measured chat usage. Instructions load when requested, subject to each skill's connection access. Recorded loads remain on individual cards and in their panels. A load means Memex sent instructions; it does not prove the assistant followed them. Your card's title and description come from its note title and saved summary. The skill panel shows that summary and links to the note editor. System and template descriptions come from files bundled with memex. **Usage example** is always visible on each user skill card and gives a prompt naming the skill by its alias, with a place for your task. The copy icon copies that prompt; it is labelled **Copy prompt**. Copying is enabled when the skill is available. For example, save your writing rules as a skill, then ask a connected assistant to use it to rewrite an update. Replace the example's brackets with the task or text you want help with. ## A skill's panel **Open skill** shows the note summary, then **Who can use this skill** with connection checkboxes. Below it, **Skill alias** and **How to use this skill** occupy two equal-width, equal-height columns. The alias has a **Save** button; the example has an embedded copy icon. The usage statistics line follows these columns. There are no expandable sections or top-right close icon; the top-right corner holds the switch. Very long titles show up to three lines in the panel; the full title remains in the note and the heading tooltip. There is one **Enable this skill** switch in the top-right corner. Turning it on makes the skill available through all supported MCP access paths (tools, resources and prompts); turning it off stops access without changing the note. Users do not choose between those technical paths. Enabling a skill makes its instructions available, but does not guarantee that an assistant will choose to use it. Connections are arranged in two equal columns. Select individual connections and press **Save connections** to limit access. Both **Save connections** and the alias field's **Save** are bordered buttons. An **Unsaved changes** label stays visible until the selected connections are saved. **All connected assistants** also includes future connections. The interface refuses to save an empty selection; turn the skill off instead. This avoids confusing an empty selection with the API's empty grant list, which means unrestricted access. The visible **Alias** field next to the chat example is the skill's exact lookup name (its slug). Most users can leave it alone. Renaming the note does not change this name, but changing the alias means older prompts using it need updating. Names are fixed when first served or opened here; collisions get a numbered suffix within 64 characters, even when another request claims a name concurrently. Pending notes have distinct temporary download names. Names accept lowercase letters, digits and single hyphens, with no whitespace. Taken names are refused without revealing the other note. Use **Edit this skill** to open the note editor. Review pending skills in the inbox. The panel also shows loads by connection, the last recorded load time, and suggestions. Sections stay visible. Full instructions are available through **Edit this skill** or **Download SKILL.md**, rather than repeated in the panel. These actions sit at the left of the footer; **Cancel** sits at the right. Cancel closes the panel and discards unfinished alias and connection selections; changes already saved, including the immediate on/off switch, remain saved. Suggestions flag missing or overly short or long descriptions, instruction bodies over about 5,000 estimated tokens, name collisions, and skills not loaded in 30 days. **Download SKILL.md** saves the portable file. ## System skills `memex-recall` explains reading and writing here; `memex-curation` contains curation rules; `memex-profile` explains profile notes; `memex-docs` is this documentation; `memex-skills` explains writing and reviewing skills; and `memex-writing` is how to write a note here, tuned by the presets you pick in **Settings › Personalization**. They update with memex and are enabled for every connection, `memex-writing` unless you switch it off. Enabled means available to load, not that their full text is inserted into every chat. ## memex-writing `memex-writing` is established practice for writing a note, drawn from style guides, standards and studies of how people and assistants read: write for a reader who was not in the conversation, absolute dates, exact identifiers, no invented reasons or sources, no chat leftovers, and the usual parts of nine kinds of note (decision, procedure, checklist, problem and fix, meeting or conversation, project status, source, concept, and person or organisation), leaving out any part the conversation did not supply. Your presets set its scope, opening, format and reasoning; see [What every setting does](https://docs.memex.tools/every-setting). Your own writing instructions, in a skill note or your profile, add to it; where the two differ it wins. Choose your own writing rules instead and it is switched off. It is an instruction to the assistant, never something memex checks, and what you ask in a conversation comes first. ## Improving one Ask your assistant: "tighten the description on my meeting-notes skill", or just point it at a finding on the page. It loads `memex-skills`, which tells it what a description an assistant can match on looks like, the size a skill should stay under, and to propose the change as a patch rather than apply it, so the edit waits in your inbox like any other. ## Your profile Your profile is a note about you, tagged `user-profile`: who you are, what you are working on, the tools you use, how you want to be answered, what to leave alone. memex names it to every assistant when it connects, and every assistant that reads it is asked to propose a small correction when a conversation shows a line has changed. It is an ordinary note: you edit it in the editor, an assistant's edit always waits in your inbox, curator or not, and deleting it or taking the tag off makes it an ordinary note again. A profile an assistant proposed is named to the others while it waits, so none proposes a second one, but it is not in force: its instructions and boundaries are not followed until you approve it. **Settings › Personalization** shows whether you have one, opens it, or starts one; or ask your assistant to start one from what it already knows about you. Keep it short: it is read often. Its sections say what you have not written yet in italics, and an assistant treats such a section as unknown. memex's own instructions for it are the built-in `memex-profile` skill, not lines in your note, so nothing you write there has to be preserved. What goes in is yours to decide: the profile's *Boundaries* section is where you say whether personal details you share may be kept in notes, and assistants apply that to every note they write here. Two profile notes are allowed; assistants are told about both and choose neither. ## Three tags memex itself reads `skill` is one. `live-state` is the second: tag a note `live-state` when it describes a system that changes (a server, a configuration, a project's current status), and every assistant that reads it over MCP is asked, in the same reply, to correct the note if its own work has just made it wrong. `user-profile` is the third, above. None of the three words can be removed from your vocabulary, because removing it would silently switch the behaviour off for every note carrying it; individual notes can gain or lose any of them freely. ## Related - [Skills](https://docs.memex.tools/skills): Notes of instructions an assistant loads when a task matches, and how to make and manage them. - [Your profile](https://docs.memex.tools/profile): A note about you that every connected assistant reads, so you don't explain yourself in every chat. - [Writing rules](https://docs.memex.tools/writing-rules): How assistants write notes in your memex, and how to tune the rules or use your own. - [How assistants receive skills and instructions](https://docs.memex.tools/skills-over-mcp): What memex tells an assistant when it connects, the ways skills are served, how a change reaches a connection, and the notices on live-state and profile notes.