How many tokens do skills and plugins use each turn?
Each turn carries only the name and description of every skill, not the full text; one plugin with 6 skills measured about 221 tokens per turn. The full skill loads when it is used. Claude Code lists what each plugin adds per turn in /usage. Codex works the same way but shows no per-skill number.
Codex
Same design: name, description and file path up front, the full skill once chosen. Codex shows no per-skill token number.
The initial skill list uses at most 2% of the context window, or 8,000 characters when the window size is unknown. With many skills installed, Codex shortens descriptions first, then leaves some skills off the list and shows a warning. Once it picks a skill, it reads the full SKILL.md. /usage in the Codex CLI shows tokens by day, by week and in total, not by skill; to carry less, disable the skills you do not use in ~/.codex/config.toml.
What the official docs say
Skills use progressive disclosure to manage context efficiently. ChatGPT and Codex start with each skill's name and description, then load the full SKILL.md instructions when they decide to use that skill. In Codex, the initial list also includes each skill's file path.To avoid crowding out the rest of the prompt, this list uses at most 2% of the model's context window, or 8,000 characters when the context window is unknown. If many skills are installed, Codex shortens skill descriptions first. For large skill sets, Codex may omit some skills from the initial list and show a warning./usage — View account token usage or use a rate-limit reset. Inspect daily, weekly, or cumulative ChatGPT token activity from inside the TUI.Claude
Only the name and description of each skill, on every turn. /usage shows what each plugin adds per turn.
Claude Code puts the names and descriptions of all skills into one listing in the system prompt. Every message carries it, used or not; after the first turn it is re-read at the cached rate. The listing is capped at 1% of the context window and one skill's description at 1,536 characters; past the cap, descriptions are dropped starting with the skills you invoke least. The full SKILL.md loads only when the skill is invoked, then stays in the conversation and is re-read on every later turn.
What the official docs say
In a regular session, skill descriptions are loaded into context so Claude knows what's available, but full skill content only loads when invoked.Every skill in the skill listing adds to your context on every turn, whether or not Claude ever uses it. Run /skill-doctor to see what each of your skills costs and how often it gets used, so you can decide which ones to turn off.The budget scales at 1% of the model's context window. When the listing overflows, Claude Code drops descriptions starting with the skills you invoke least, so the skills you use most keep their full text.Skills load on-demand only when invoked, so moving specialized instructions into skills keeps your base context smaller.The plugin row in /usage
Run /usage in Claude Code and find the row named Plugin skill-listing footprint. It lists each plugin with its skill count and its cost per turn, for example: diagram-design 6 skills · ~221 tok/turn. That figure is one measured plugin, not an average.
The /usage panel describes the row this way: “What each plugin's skill descriptions add to the system prompt (cached input after the first turn). Agents and MCP tools not yet counted.” This sentence comes from the panel, not from the docs. Agents and MCP tools that a plugin ships come on top of the number shown.
Turn off the skills you do not use
A skill that is turned off stops adding its description to every turn. How you turn it off depends on where the skill came from.
Claude Code, skills from a plugin: run /plugin disable plugin-name@marketplace-name, or use the Installed tab in /plugin. Plugin skills can only be turned off this way.
Claude Code, your own skills: open /skills, highlight a skill, press Space to change its state, then Esc to save. name-only keeps just the name; off hides the skill completely.
Claude Code, skills you only start by hand: add disable-model-invocation: true to the top of SKILL.md. Its description is no longer carried on each turn.
Codex: add a [[skills.config]] entry to ~/.codex/config.toml with path = "/path/to/skill/SKILL.md" and enabled = false, then restart Codex.
What the official docs say
Plugin skills are not affected by skillOverrides. Manage those through /plugin instead.Use [[skills.config]] entries in ~/.codex/config.toml to disable a skill without deleting itCommon misconceptions
- With 50 skills installed, every turn reads 50 full skill files
No. Each turn carries names and descriptions only; 6 skills measured about 221 tokens in total. The full text loads only when that skill is used.
- The more detailed a skill description, the better
No. The description is the part carried on every turn, and it gets cut: Claude Code allows 1,536 characters per skill and 1% of the context window for the whole listing; Codex allows 2% for its list. Put the main use and the trigger words first, and keep it short.
- Once a skill has run, it stops using tokens
No. The full skill stays in the conversation after it loads and is re-read on every later turn. After Claude Code compacts a session, each invoked skill keeps its first 5,000 tokens, up to 25,000 in total. Start a new session when you change tasks.
Tips
- Claude Code: run /usage and read the Plugin skill-listing footprint row. Start with the plugin that adds the most tokens per turn.
- Claude Code: run /skill-doctor (v2.1.252 or later). It marks skills that were never invoked and plugins you have not used recently. The Skills line in /context is the current size of the whole listing.
- After a skill has run and you move to another task, use /clear (/new in Codex). The full skill text stays in the session until you do.