For developers and founders

Build faster with the right skills for the job.

Skillstore is a searchable skills library for Claude Code and Codex. Connect through MCP to find expert instructions for what you’re building. Add your team’s own methods so teammates and coding agents can follow the same standards.

Try a review in Claude Code · Try a review in Codex

No Skillstore account needed for the public library. Use your existing AI account.

What changes in my work?

A skill is a reusable set of instructions for a specific job: how to structure a SwiftUI view, investigate a slow query, or check access to an API resource. Your agent reads the skill and does the work.

In a real Claude Code run on a fictional invoice API, Skillstore delivered Addy Osmani and contributors’ Security and hardening skill. The review found that a signed-in user could read another organization’s invoice. The fix checks the resource’s organization; four checks passed after the fix. Inspect the code and verification.

Public skills behind these examples:

Explore the engineering workflow: code quality, performance and team standards

Can I try it now?

Connect the public library and start a fresh session. Try the unsolved invoice API with failing checks, with no private code needed:

Review the fictional invoice endpoint at https://skillstore-jk.fly.dev/examples/invoice-task.mjs in a scratch folder. Read the file before running it. Use Skillstore to find relevant security-review guidance, sending only a generic task description and necessary technical context. Read the method that fits and any required supporting files through the supplied file getters, then apply the relevant guidance; if none fits or the connection is unavailable, say so and continue with your usual review. Run node --test invoice-task.mjs to see the failing checks, fix the endpoint without changing the tests, then rerun the same command. Explain the bug, name the method you actually read, the criterion you applied and the finding it supports. Separate method guidance from your own analysis and report both test results. If your tools cannot run the checks, say so.

Use your own change instead

Review this change. Read the repo’s instructions and relevant build files first. Use Skillstore to find guidance for the task and stack, sharing only a generic task description and necessary technical context. Read the method and any required supporting files through the supplied file getters, then apply relevant checks. Name the method you actually read, the criterion you applied and the finding it supports. Separate method guidance from your own analysis, and report what you verified. If nothing fits or the connection is unavailable, say so and continue with your usual review.

Run the same checks before and after. Ask your agent to show the test results and explain which method helped. After the fix, compare with the worked reference.

Plain-text setup and disconnect guide

Why not just use AGENTS.md or GitHub?

Find the right approach, faster. Skillstore finds relevant skills for the task and the tools your project uses, then delivers the selected instructions and supporting files through MCP. Keep project rules in AGENTS.md and skills you prefer to use locally. Skillstore adds a searchable library and controlled access to private methods. GitHub imports are snapshots; upstream changes need to be imported again.

Can I bring my own methods?

Turn your best work into your team’s standard. Ask your AI to turn a task you’ve completed into a skill, or import instructions you already use. Save the skill privately and share it with your team. The next developer starts with the same method. A Skillstore account is needed for your private library.

Skillstore learns what works for your project. Tell Skillstore what helped, what missed, and what your team does differently. Use that feedback to improve skill selection and save corrections for the next task.

Add your own skills

What reaches Skillstore?

Skillstore receives the arguments your agent sends in tool calls. Connecting the library does not automatically give Skillstore your repository or full conversation. Keep secrets out of task descriptions. Private skills require an authorized connection. You can invite one person or share a link with several people. Each recipient signs in with their own account. Keep invitation and connector links private. Revoking access stops future reads through that grant; it cannot remove instructions already delivered or copied.

How does Skillstore learn what works?

Feedback about useful and poor matches helps improve skill selection. In your private library, ask your agent to remember a correction to a skill. Review and approve the rule and where it should apply before Skillstore saves it.

For example: “For this project, always check organization access before returning an invoice.” On a later matching task, your agent can retrieve that saved instruction. You choose what becomes a reusable rule.