Use case · Skills

# Our best agent skills live on one laptop. *Give each engineer the same catalog on every machine.*

Keep your team's skills in one GitHub repo. Each engineer imports it once, and the skill that matches loads on every machine they use. Import again to update.

[Get started free](https://console.context-mode.com/login) [Read the docs](https://context-mode.com/docs/skills#import)

Pick a moment

One laptop Import Teammate Plain words Just a question Right pick

The team's best skills sit in one home folder.

Built into each agent

- SKILL.md files on each machine or repo
- Copied by commit, plugin or admin folder
- A folder for each agent

One catalog on your account

- Imported from your team's GitHub repo
- One catalog for both agents
- Nothing written to the disk

Each machine or repo needs its own copy. Context Mode keeps the skills on the gateway.

Put the skills in a GitHub repo. Import it once.

```
// console → Skills
https://github.com/your-org/agent-skills
→ Preview import
→ Add all
```

Large repos import in batches. Import again to update.

A teammate imports the same repo on their account.

```
// their console → Skills
https://github.com/your-org/agent-skills
→ Preview import → Add all
```

Their skills *diagnosing-bugs* *to-tickets* *batch-grill-me*

Same repo, same skills. When the repo changes, each of you imports again.

“Since this morning our blog URLs end with a dash. … the test suite is red. … Figure out what broke and fix it.”

Your skills *diagnosing-bugs* *to-tickets* *batch-grill-me*

```
⏺ context-mode · skill diagnosing-bugs auto-loaded
```

Nobody typed a skill name. The reply names the skill that loaded.

“Walk me through how slugify turns a title into a slug, step by step.”

Your skills *diagnosing-bugs* *to-tickets* *batch-grill-me*

This question matches no skill, so none loads. If the match is weak, no skill loads.

Right skill picks on 10 texts, offline.

Before **4 of 10** Now, short list **9 of 10** Now, all skills **10 of 10**

Small sample, judged by one person. Treat it as direction.

The problem

## One engineer wrote the skills. *The rest of the team never gets them.*

**Skills are files on each disk**

The agent loads skills from `SKILL.md` files. Each machine or repo needs its own copy, by commit, plugin or admin folder.

**Each agent has its own folder**

Claude Code and Codex read skills from different folders. A team that uses both keeps a copy for each.

**The model picks on its own**

In each agent, the model loads a skill when its description fits, using only the files on that machine.

How it works

## From a repo to every agent. *No files to sync.*

1. 01 **Put the skills in a GitHub repo** One folder per skill, each with a `SKILL.md` and a clear description.
2. 02 **Each engineer imports it once** In the console, open Skills, paste the repo URL, click Preview import, then Add all. The catalog is one per account, and there are no team roles yet, so each engineer imports the same repo. Import again to update. A skill body can be 262,144 bytes at most.
3. 03 **The right skill loads** On every machine you use, Context Mode compares each message with each skill. At most one loads per turn, and the reply names it.

Agents on any laptop send each message through Context Mode. It holds the catalog imported from GitHub, picks one skill or none, adds it to the request and names it in the reply. Then the request goes to the model on your own account. Nothing is written to the disk.

Proof

## The right skill, or none. *Named in every reply.*

**10 of 10** right skill picks on a small offline test, against 4 of 10 with the old judge

**About 760** tokens a request for the skill list in our benchmark accounts, read from cache

|  | Context Mode | Built into each agent |
| --- | --- | --- |
| Where skills live | One catalog on your account | Files on each machine or repo |
| New laptop | Ready at once on your account; each teammate imports the repo once | Commit, plugin or admin folder on each machine |
| Claude Code and Codex | One catalog for both | A folder for each agent |
| Which skill loads | The one that matches, named in the reply | The agent decides |

Skill picks: an offline eval on 10 texts, judged by one person. Treat the numbers as direction, not as live accuracy. We have not measured whether skills give better answers. The catalog is one per account; there are no team roles yet, so each engineer imports the same repo, and imports again after a change. [Method and sources →](https://context-mode.com/blog/the-right-skill-at-the-right-time#proof)

Products used

## What this *uses.*

**[Skills](https://context-mode.com/skills)**

Import from GitHub once per account. The skill that matches loads, one per turn at most.

**[Skills docs](https://context-mode.com/docs/skills)**

Import, matching, the reply line, cost, and a 5-minute test.

**[How a skill is picked](https://context-mode.com/blog/the-right-skill-at-the-right-time)**

The judge, the 0.5 floor, and what we have not measured.

**Related**

[Team standards](https://context-mode.com/solutions/team-standards) · [For platform teams](https://context-mode.com/solutions/platform-teams) · [Instructions](https://context-mode.com/docs/instructions)

## Write a skill once. *Every teammate uses it.*

*npx @context-mode/cli* connects your agents. 1,000 requests free, no card.

[Get started free →](https://console.context-mode.com/login) [Read the docs →](https://context-mode.com/docs/skills)