Skip to content

skilldrop·docs·How-to

Use the API builder pack

Install the API builder pack, run its first task, and follow a typical session step by step.

Skills for teams building on the Anthropic API — prompt caching, token budgeting, eval generation for skills, and tool-use schema writing.

Reach for it when: You are building or tuning a Claude API integration and want its tools defined properly and its token costs under control.

1. Install it

npx skilldrop-cli install --pack api-builder                 # Claude Code, your user folder
npx skilldrop-cli install --pack api-builder --project       # Claude Code, this repo
npx skilldrop-cli install --pack api-builder --ide cursor    # or kiro, codex, antigravity, copilot
npx skilldrop-cli install --pack api-builder --local         # this repo, hidden from git

In Claude Code you can also add the marketplace once (/plugin marketplace add sananthanarayan/skilldrop) and run /plugin install api-builder@skilldrop.

2. Run the first task: find where prompt caching will cut your input costs

Before you start:

Paste this into your agent:

Here is our system prompt and a typical request: <paste them>. Where should the cache_control breakpoints go, and what hit rate should we expect?

How to tell it worked: prompt-caching-advisor marks where each cache_control checkpoint goes and estimates the cache hit rate.

If nothing happens: If prompt-caching-advisor does not activate, ask for it by name and check that its folder exists in your skills folder (~/.claude/skills/ by default).

3. A typical session

Bring:

A session takes about 30–60 minutes.

  1. tool-use-schema-writer: Describe each function the model should call. You get: A tool_use JSON schema per function with types, required fields and parameter descriptions, and a Python client.messages.create snippet when you ask for one.
  2. token-budget-estimator: Describe the workflow or share a sample request and response. You get: A per-component token table, the dominant cost driver, and max_tokens and other budget controls per component.
  3. prompt-caching-advisor: Paste the system prompt and a typical request. You get: Each block classed as static, quasi-static or dynamic, where each cache_control breakpoint goes with code, and an expected cache hit rate.

You end with: Tool schemas ready to wire in, a token budget with its main cost driver named, and a caching plan.

Next

Edit this page on GitHub · View source