Stop Pasting API Docs Into Your AI Agent: Generate Reusable Skills Instead

Alex — Fri, 29 May 2026 03:02:31 +0000

AI coding agents are powerful, but they still waste a lot of context on the same setup work:

reading a huge OpenAPI document again
rediscovering GraphQL queries and variables
figuring out MCP tool names from scratch
remembering which auth header or environment variable a service expects

@asnd/skill-creator is a small TypeScript CLI that turns those API surfaces into reusable agent skills and wrapper scripts.

Instead of pasting API docs into every chat, you install one /skill-creator command, point it at an OpenAPI spec, GraphQL endpoint/schema, or MCP server, and let your agent create a focused skill that future agents can reuse.

The idea

Good agent workflows are repeatable. If an agent learns how to use your internal API once, that knowledge should become a durable tool, not disappear when the chat ends.

skill-creator helps create that durable layer:

npx @asnd/skill-creator command install --agent pi --scope global

Then, inside the agent:

/skill-creator https://example.com/openapi.json

The agent researches the source, generates a skill, writes usage notes, saves the source reference, and smoke-tests the wrapper.

What it creates

A generated skill looks like this:

.pi/skills/youtube/
├── SKILL.md
├── scripts/
│   └── youtube
└── references/
    └── openapi-spec-MM-DD-YYYY.json

Future agents can discover and call operations through a stable CLI:

./scripts/youtube commands list
./scripts/youtube commands search videos
./scripts/youtube commands help <command>
./scripts/youtube run --pretty <command> <flags>

That means the next agent does not need to re-read the whole API spec. It starts with focused commands, examples, gotchas, and safe usage rules.

Supported sources

skill-creator currently supports:

OpenAPI specs, local or remote
GraphQL endpoints with SDL or introspection JSON schemas
MCP servers over HTTP/SSE
MCP stdio servers, for example npx -y some-mcp-server

Examples:

npx @asnd/skill-creator generate \
  --template openapi \
  --name youtube \
  --spec https://example.com/openapi.json \
  --agent pi \
  --scope project

npx @asnd/skill-creator generate \
  --template graphql \
  --name pokeapi \
  --graphql https://api.example.com/graphql \
  --graphql-schema ./schema.graphql \
  --agent pi \
  --scope project

npx @asnd/skill-creator generate \
  --template mcp-stdio \
  --name example-mcp \
  --mcp-stdio "npx -y @example/mcp-server" \
  --agent pi \
  --scope project

Why this is useful

The main benefit is not code generation. The generated wrapper delegates to npx -y @asnd/skill-creator at runtime, so consumers do not need a global install and the command surface stays dynamic.

The useful part is the agent-facing workflow around it:

commands list, commands search, and commands help make tools discoverable.
Source specs and schemas are saved under references/ for reproducibility.
Auth can stay in environment variables or files instead of copied into docs.
Generated SKILL.md files capture gotchas, safe usage patterns, and verified examples.
A companion improvement skill can update generated skills when future agents learn reusable quirks.

Multi-agent support

The install command supports several coding agents, including:

Pi
Claude Code
Codex
Cursor
opencode
Gemini CLI
GitHub Copilot
Cline
Windsurf

For example:

npx @asnd/skill-creator command install --agent claude-code --scope project

Try it

The package is available on npm:

npx @asnd/skill-creator --version

Repository:

https://github.com/sandiiarov/skill-creator

If you are building with AI agents and repeatedly paste the same API docs into chats, skill-creator is a way to turn that repeated context into a reusable command-line skill.