SkillFlux

    ↑ ↓ select↵ openSearch all of SkillFlux
    Get started

    EQUIP YOUR AGENT

    Give your AI agent new capabilities.

    Equip Claude, Cursor, and Codex with the exact capabilities your project demands. Search, review, and load on demand.

    Node.js ≥ 22.13.0npm + MCPAvailable: source install
    Before you start

    Source and local build packages are available. Run the first commands from your downloaded SkillFlux repository root; a public npm release is not assumed.

    01

    Install the local CLI

    Build the current version and install the skillflux command. Check your Node.js version, then run these four commands.

    TERMINAL · REPOSITORY ROOT
    npm ci
    npm run build --workspace @skillflux/mcp
    npm pack --workspace @skillflux/mcp
    npm install -g ./skillflux-mcp-1.0.0.tgz
    02

    Connect the skill catalog

    SkillFlux curated skills live in a GitHub catalog repository. init resolves the repository's current commit and pins it; every download then anchors to that commit and verifies per-file hashes.

    vc999999999/skillflux-catalog

    Only evaluated and approved releases can be installed. Check the curated registry or search with the client for available versions.

    03

    Choose your AI tool

    Open your target project, choose a host and copy the command. init merges the project's MCP configuration, preserves existing servers and pins the catalog repository identity.

    TERMINAL · YOUR PROJECT
    cd /path/to/your-project
    skillflux init --project . --repo vc999999999/skillflux-catalog --host codex

    MCP config.codex/config.toml

    Entry skill.agents/skills/skillflux/

    Replace /path/to/your-project with your real project path. Reload MCP configuration, then explicitly invoke SkillFlux. Use $skillflux in Codex; generic hosts may need the entry loaded manually.

    Allow MCP to install reviewed text bundles

    Only an explicit --preauthorize-reviewed-text flag on init allows MCP to install approved text bundles with completed purpose and boundary tests and no network, shell or secret capabilities. CLI installation is the default; exact-version upgrades still need approval.

    04

    Search, review, install

    Search runs locally over a cached catalog index; queries never leave your machine. Choose a skill and exact version, review the plan's capabilities, dependencies and files, then install using its returned plan ID.

    TERMINAL · YOUR PROJECT
    skillflux search "code review"
    skillflux plan SKILL_ID --version VERSION
    skillflux install PLAN_ID
    skillflux load SKILL_ID

    Replace SKILL_ID, VERSION and PLAN_ID with actual values. Installation downloads each file at the pinned commit and rechecks hashes, qualification and revocations. load provides the skill entry and needed resources in the current turn.

    05

    You decide when to update

    check-updates is read-only. update with an exact version shows a plan and waits for terminal confirmation. pin locks a version; unpin it before upgrading.

    TERMINAL
    skillflux check-updates
    skillflux update SKILL_ID@VERSION
    skillflux pin SKILL_ID

    The catalog index cache lasts up to 24 hours; manual checks refresh by default. Offline results show cache age and unknown states. Installation, upgrades and rollback require online qualification and revocation checks.

    06

    Common questions

    Can I install every directory source?

    The source directory helps you discover third-party resources. SkillFlux installation commands apply to approved versions in your catalog repository. Follow the original instructions for other MCP servers and collections.

    Does setup modify global AI settings?

    init writes only the specified project's MCP configuration and entry files, preserves existing servers and does not overwrite user-owned entry files.

    Will every skill run automatically?

    An explicit user invocation is required. After reading the lightweight entry, the host uses the local MCP runtime to search, verify and load. The website does not run your model.

    Connected. Find the right skill next.

    Explore available versions or start with the source directory.

    Browse reviewed skills