skill-kit
Agent skills for the points where an agent's work reaches a person.
| When | Skill | What it does |
|---|---|---|
| A decision is needed | quiz | Asks it as multiple choice |
| Options to pick from | variant-lab | Builds and tests 4 to 6 working versions |
| Writing is handed over | writing-check | Flags AI writing patterns |
| A graded or shared document | deliverable | Drafts it from the rubric |
| The agent says done | proof-before-done | Shows the tool result as proof |
| A push to a public repo | push-check | Blocks secrets, malicious code and personal information |
| A session or tool ends | handoff | Writes and picks up handoff notes |
These are the points where agent work costs a person time or embarrassment: a question buried in a long reply, a claim with no source, a key in a public commit. Process kits such as superpowers cover how an agent builds. This kit covers what happens when that work meets a person, and works alongside them.
Each skill is a plain SKILL.md folder, so any agent tool that reads that format can use it.
Install
Claude Code
/plugin marketplace add S3yon/skill-kit
/plugin install skill-kit@skill-kit
The skills then show as skill-kit:<name>, for example skill-kit:quiz.
Checked against https://code.claude.com/docs/en/plugin-marketplaces, 2026-10-02.
GitHub Copilot CLI
Reads the same plugin manifests as Claude Code.
copilot plugin marketplace add S3yon/skill-kit
copilot plugin install skill-kit@skill-kit
Or in one step: copilot plugin install S3yon/skill-kit.
Checked against https://docs.github.com/en/copilot/reference/copilot-cli-reference/cli-plugin-reference, 2026-10-02.
Qwen Code
Reads the same marketplace file as Claude Code.
qwen extensions install S3yon/skill-kit:skill-kit
Checked against https://github.com/QwenLM/qwen-code/blob/main/docs/users/extension/introduction.md, 2026-10-02.
Gemini CLI
Reads gemini-extension.json and loads skills/.
gemini extensions install https://github.com/S3yon/skill-kit
Checked against https://github.com/google-gemini/gemini-cli/blob/main/docs/extensions/reference.md, 2026-10-02.
Codex
Reads the root plugin.json and finds skills/ on its own.
codex plugin marketplace add S3yon/skill-kit
codex plugin add skill-kit@skill-kit
Checked against https://developers.openai.com/plugins/build/plugins, 2026-10-02.
Cursor
Reads the root plugin.json and .cursor-plugin/marketplace.json. Open Customize, then
From GitHub Repository, and enter S3yon/skill-kit.
Checked against https://cursor.com/docs/plugins, 2026-10-02.
Copy a folder
Copy one folder from skills/ into the tool's skills folder:
| Tool | For your user | For one project |
|---|---|---|
| Claude Code | ~/.claude/skills/ | .claude/skills/ |
| Codex, Cursor, Gemini CLI, Copilot CLI | ~/.agents/skills/ | .agents/skills/ |
| Qwen Code | ~/.qwen/skills/ | .qwen/skills/ |
No manual install test was run. Of these tools, only the Cursor CLI was installed on the machine where the steps were checked, and it was not signed in, so it could not load a plugin.
Always-on snippet
Skills load when their description matches the task. To have three of the rules apply on every task, paste these lines into a project's agent instructions file:
- Report an action as done only with the tool result that shows it, and an outside fact only with its source.
- Put any draft that needs approval in full in the reply text.
- Run push-check before any push to a public remote.
The scripts on their own
Both need only Python 3.10 or later. push_check.py also needs git.
skills/writing-check/scripts/slop_check.py checks a draft:
python3 skills/writing-check/scripts/slop_check.py draft.md
It prints one line per flag with the line number and a replacement. Exit 0 means clean, 1 means something was flagged. The patterns and why they are on the list: tells.md.
skills/push-check/scripts/push_check.py checks commits before a push:
python3 skills/push-check/scripts/push_check.py --remote origin
python3 skills/push-check/scripts/push_check.py --install-hook
With no range it scans the unpushed commits. --install-hook installs it as the repo's
pre-push hook. Exit 0 means nothing blocks, 1 means something does, 2 means a usage error.
Tests, CI and evals
python3 -m unittest discover testsruns the script tests.python3 scripts/validate.pychecks every skill folder, eval file and plugin manifest.- CI runs both on every push and pull request, on Python 3.10 and the latest 3.x.
- Each skill's
evals/evals.jsonholds prompts with checkable assertions. How to run and grade them: docs/evals.md.
License
MIT