UXC Skill

Use this skill when a task requires calling a remote interface and the endpoint can expose machine-readable schema metadata.

When To Use

• You need to call APIs/tools from another skill and want one consistent CLI workflow.
• The interface may be OpenAPI, GraphQL, gRPC reflection, MCP, or JSON-RPC/OpenRPC.
• You need deterministic, machine-readable output (ok, kind, data, error).

Do not use this skill for pure local file operations with no remote interface.

Docs Search & Support

• UXC docs support full-text search at https://uxc.holon.run/api/search?q=<query>.
• Prefer searching docs first when operation names, auth flags, or protocol behavior are unclear.
• If docs are unclear or behavior looks wrong, open an issue in holon-run/uxc:
  • https://github.com/holon-run/uxc/issues/new/choose
  • include command, endpoint, and the JSON envelope (ok, error, meta) for faster triage.

Prerequisites

• uxc is installed and available in PATH.
• For gRPC runtime calls, grpcurl is installed and available in PATH.

Install uxc

Choose one of the following methods:

Homebrew (macOS/Linux):

brew tap holon-run/homebrew-tap
brew install uxc

Install Script (macOS/Linux, review before running):

curl -fsSL https://raw.githubusercontent.com/holon-run/uxc/main/scripts/install.sh -o install-uxc.sh
# Review the script before running it
less install-uxc.sh
bash install-uxc.sh

Cargo:

cargo install uxc

For more options, see the Installation section in the UXC README.

Core Workflow

1. Discover operations:
   • uxc <host> -h
2. Inspect a specific operation:
   • uxc <host> <operation> -h
3. Execute with structured input:
   • uxc <host> <operation> key=value
   • uxc <host> <operation> '<payload-json>'
4. Parse result as JSON envelope:
   • Success: .ok == true, consume .data
   • Failure: .ok == false, inspect .error.code and .error.message
5. For disambiguation, use operation-level help first:
   • uxc <host> <operation> -h
6. For auth-protected endpoints, use the right auth track:
   • simple bearer / single-secret API key: see references/auth-configuration.md
   • multi-field auth or request signing: see references/auth-configuration.md
   • OAuth flows: see references/oauth-and-binding.md

Link-First Workflow For Wrapper Skills

Wrapper skills should default to a fixed local link command instead of calling uxc <host> ... directly on every step.

1. Pick a fixed command name during skill development:
   • naming convention: <provider>-mcp-cli
   • examples: notion-mcp-cli, context7-mcp-cli, deepwiki-mcp-cli
2. Check whether the command already exists:
   • command -v <link_name>
3. If command is missing, create it:
   • uxc link <link_name> <host>
   • For OpenAPI services whose schema is hosted at a separate fixed URL, create the link with uxc link <link_name> <host> --schema-url <schema_url>
   • For stdio hosts that need credential-driven child env auth, create the link with uxc link <link_name> <host> --credential <credential_id> --inject-env NAME={{secret}}
   • If the link is being created as part of a wrapper skill, persist source metadata with --skill <skill_name> --skill-doc <docs_url> --skill-path <local_skill_path> so later help output preserves skill context.
4. Validate link command:
   • <link_name> -h
5. Use only the link command for the rest of the skill flow.

Import Existing MCP Config First

If an MCP server is already configured in a supported editor or local agent, prefer importing it before hand-writing a new link:

• Preview imports:
  • uxc config import mcp --dry-run
• Auto-discover common MCP config sources and import them:
  • uxc config import mcp --from auto
• Import from a specific source preset:
  • uxc config import mcp --from cursor
  • uxc config import mcp --from codex

Supported presets in v1 include:

• auto
• cursor
• claude-code
• claude-desktop
• vscode
• codex
• windsurf
• opencode

Naming Governance

• Link naming is a skill author decision, not a runtime agent decision.
• Resolve ecosystem conflicts during skill development/review.
• Do not implement dynamic rename logic inside runtime skill flow.
• If runtime detects a command conflict that cannot be safely reused, stop and ask for skill maintainer intervention.

Equivalence Rule

• <link_name> <operation> ... is equivalent to uxc <host> <operation> ....
• If the link was created with --schema-url <schema_url>, it is equivalent to uxc <host> --schema-url <schema_url> <operation> ....
• If the link was created with --credential <credential_id> --inject-env NAME={{secret}}, it is equivalent to uxc --auth <credential_id> --inject-env NAME={{secret}} <host> <operation> ....
• Callers can still override that persisted schema by passing --schema-url <other_url> explicitly at runtime.
• Use uxc <host> ... only as a temporary fallback when link setup is unavailable.

Input Modes

• Preferred (simple payload): key/value
  • uxc <host> <operation> field=value
• Bare JSON positional:
  • uxc <host> <operation> '{"field":"value"}' Do not pass raw JSON through --args; use positional JSON.

Output Contract For Reuse

Other skills should treat this skill as the interface execution layer and consume only the stable envelope:

• Success fields: ok, kind, protocol, endpoint, operation, data, meta
• Failure fields: ok, error.code, error.message, meta

Default output is JSON. Do not use --text in agent automation paths.

Reuse Rule For Other Skills

• If a skill needs remote API/tool execution, reuse this skill instead of embedding protocol-specific calling logic.
• Wrapper skills should adopt a fixed link command (<provider>-mcp-cli) as the default invocation path.
• Upstream skill inputs should be limited to:
  • target host
  • operation id/name
  • JSON payload
  • required fields to extract from .data

Reference Files (Load On Demand)

• Workflow details and progressive invocation patterns:
  • references/usage-patterns.md
• Generated runtime client flow:
  • https://uxc.holon.run/ecosystem/typescript-client/
• Protocol operation naming quick reference:
  • references/protocol-cheatsheet.md
• Public endpoint examples and availability notes:
  • references/public-endpoints.md
• Authentication configuration (simple secret, named fields, headers/query params, and request signers):
  • references/auth-configuration.md
• OAuth and credential/binding lifecycle:
  • references/oauth-and-binding.md
• Failure handling and retry strategy:
  • references/error-handling.md