Guide · Agent skill
How to use the Extract Patterns into Skills agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The extract skill, published in the alirezarezvani/claude-skills repository under engineering-team/self-improving-agent, turns a proven pattern or debugging solution into a standalone skill folder. It is written for developers and teams who keep solving the same environment or build problem across projects and want that fix packaged for reuse. After this guide you will know what the skill does, how to install it, how to trigger it with a natural request or the /si:extract command, and what to check before you trust the output.
Extract Patterns into Skillsalirezarezvani/claude-skills · MIT Download the skill Quick start
- 1Download the ZIP. Download the extract skill ZIP from the skill page on this site.
- 2Install the skill. Unzip it into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder, or upload the ZIP under Settings → Capabilities → Skills in the Claude apps.
- 3Check the requirements. Confirm your agent runtime reads SKILL.md files, the self-improving-agent skill set and skill-extractor agent are available, and Bash access is allowed.
- 4Preview with dry-run. Run /si:extract with your pattern and --dry-run to see the planned output without creating files.
- 5Extract the pattern. Run /si:extract with a clear description, optionally adding --name and --output, and answer the agent's up to two questions.
- 6Review the generated folder. Read the SKILL.md, README.md and any reference/examples.md, check for project-specific values and confirm the fix works.
- 7Reuse the folder. Copy the skill folder into another project or skills directory where you need the same fix.
What the extract skill makes the agent do
The extract skill is a set of instructions that tells the agent how to package a solution it has already found. The source describes it as turning a recurring pattern or debugging solution into a standalone, portable skill that can be installed in any project.
The agent first looks for related entries in auto-memory, then asks up to two clarifying questions if needed, generates a skill name, and hands the file creation to a separate skill-extractor agent. The output is a folder containing a SKILL.md, a README.md and an optional reference/examples.md.
- Searches auto-memory notes for matching entries and uses them as source material
- Asks at most two questions about the problem or whether code examples should be included
- Creates a folder named with lowercase hyphenated words, such as docker-m1-fixes
- Runs a checklist on the result before reporting the files it created
How to install the extract skill
Download the ZIP file from the skill page on this site. Then choose one of the two install routes below, depending on how you work with your agent.
For Claude Code, unzip the file into ~/.claude/skills/ to make it available everywhere, or into a project's .claude/skills/ folder to limit it to that project. In the Claude apps, you can instead upload the ZIP under Settings → Capabilities → Skills.
- Personal install: unzip into ~/.claude/skills/
- Project install: unzip into the project's .claude/skills/ folder
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
How to trigger the extract skill with a natural request
You can use the explicit command /si:extract followed by a description of the pattern. The skill's description also says it applies when a user asks to package a recurring solution from memory into a skill, so a plain request such as asking the agent to save a fix as a skill can trigger it.
The source lists user-flagged phrases such as 'Save this as a skill' and 'I want to reuse this' as signals that extraction is wanted. Give the agent the problem, the fix and the context, because the output is only as good as the description.
- Explicit: /si:extract with a pattern description
- Natural: ask the agent to package a solution it has learned into a reusable skill
- Useful phrasing: say what broke, what fixed it and where else you expect it to recur
/si:extract <pattern description>
/si:extract <pattern> --name docker-m1-fixes
/si:extract <pattern> --output ./skills/
/si:extract <pattern> --dry-runOptions and worked examples
The --name flag sets the skill name, --output sets the output directory, and --dry-run previews the result without creating files. The source gives two examples: a Docker build failure on Apple Silicon, which would produce a docker-m1-fixes folder with the error message, three solutions, a trade-offs table and a performance note, and a workflow rule about regenerating a TypeScript API client after an OpenAPI change, which would produce an api-client-regen folder.
Names must be lowercase with hyphens, 2-4 words, and must not contain the reserved fragments claude or anthropic. For skills about Claude Code itself, the source asks for a cc- prefix instead, for example cc-settings.
- Use --dry-run first to see what the agent would write
- Names like docker-m1-fixes or pnpm-workspace-setup follow the source convention
- A reserved word in the name is changed and confirmed with you before files are written
/si:extract "Fix for Docker builds failing on Apple Silicon with platform mismatch"What you need before using it
The source lists these requirements: a SKILL.md-compatible agent runtime, the self-improving-agent skill set with a skill-extractor agent, and Bash access so the agent can search the auto-memory directory under $HOME/.claude/projects. The skill runs scripts, and the source states it does not use the network.
The auto-memory search is a grep over the memory folder for your project. If your setup has no memory notes, the agent falls back to your description alone.
- Compatible agent runtime that reads SKILL.md files
- Self-improving-agent skill set and the skill-extractor agent
- Bash access to search $HOME/.claude/projects memory folders
- Runs scripts: yes. Uses network: no
MEMORY_DIR="$HOME/.claude/projects/$(pwd | sed 's|/|%2F|g; s|%2F|/|; s|^/||')/memory"
grep -rni "<keywords>" "$MEMORY_DIR/"Safety notes before you reuse an extracted skill
The extracted skill copies whatever the agent believes works. If the fix was wrong, the wrong fix is packaged and reused, so verify it against a real run before sharing. The source's quality checklist asks for no project-specific hardcoded paths, URLs or credentials, but you should still read the output yourself.
Because the skill runs scripts and searches your memory directory with Bash, review the commands it proposes and point it only at the project you intend to search.
- Test the fix yourself before extracting it
- Read the generated SKILL.md for project-specific paths, URLs or credentials
- Check that the solution works without the original context
- Remember the source says the skill is not meant for one-off, single-codebase fixes
Tips for getting useful skills
Extract patterns that would save time in a different project, and keep each skill focused on one problem. Include the error messages people would search for, because the generated skill is meant to be found by its symptoms. The source also suggests testing the result by reading it without the original context to see whether it still makes sense.
- Pick patterns that recur across two or more projects
- Keep one problem per skill
- Include exact error text in the description
- Reread the output as if you had never seen the original bug
What people use it for
- Packaging a Docker platform-mismatch fix on Apple Silicon so it can be reused on other machines
- Capturing a multi-step build or environment fix that is easy to forget
- Turning a regenerate-after-change workflow, such as an API client rebuild, into a documented skill folder
- Sharing debugging know-how across a team as installable skill folders
Frequently asked questions
Is the extract agent skill free to use?
The source page states the skill is released under the MIT licence. Check the repository's licence file for the full terms before redistributing it.
What do I need to run the extract skill?
You need a SKILL.md-compatible agent runtime, the self-improving-agent skill set with a skill-extractor agent, and Bash access to search your auto-memory directory under $HOME/.claude/projects. The source says it runs scripts and does not use the network.
Where do I install the extract skill?
Unzip the downloaded ZIP into ~/.claude/skills/ for all projects or into a project's .claude/skills/ folder for one project. In the Claude apps you can upload the ZIP under Settings → Capabilities → Skills.
How do I trigger the extract skill?
Run /si:extract followed by a pattern description, or ask the agent in plain language to package a recurring solution from memory into a skill. Use --dry-run first if you want to preview the output.
Can the extract skill package a wrong fix?
Yes. The source notes the generated skill copies whatever the agent believes works, so an incorrect fix would be packaged and reused. Test the solution before you extract it and review the output.
Is extract a good alternative to writing skills by hand?
It suits repeated fixes you already recorded in memory notes, because it searches those notes and fills the standard SKILL.md structure. For a one-off fix tied to one codebase, the source says this skill is not the right fit, and writing a short note by hand may be simpler.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
The extract skill is worth trying if you already keep auto-memory notes and repeatedly solve the same environment or build problem across projects. Skip it for one-off fixes tied to a single codebase, or if you cannot set up the self-improving-agent skill set and skill-extractor agent it depends on.
Download the skill