Teach your coding agent to own API workflows with skills
Divyank Jain describes how to stop manually running API calls by packaging reusable "skills" — a `SKILL.md` file plus a Python CLI — that coding agents like Claude Code can load on demand to execute API workflows autonomously.
Score breakdown
Teams using multiple coding agents (Claude Code, Cursor, Copilot, etc.) can generate one reusable API skill and share it across all of them, eliminating per-session doc re-reading and manual endpoint wrangling for every developer.
- 01A "skill" is a folder containing a `SKILL.md` file that packages API instructions, auth context, and an optional Python CLI script for a coding agent to load on demand.
- 02`api-skill-creator` generates a complete skill (SKILL.md + Python CLI) from OpenAPI 3.x, Swagger 2.0, Postman collections, or HTML docs pages with a single command and no external API calls.
- 03The tool has only one dependency (`pyyaml`) and works fully offline.
Divyank Jain describes a frustration common to developers using coding agents: even with a capable agent like Claude Code handling code browsing, test writing, and command execution, the developer is still manually calling API endpoints, copying values between requests, and managing auth. The root problem, he argues, is that agents re-read the same API docs, re-discover the same endpoints, and re-reason about the same auth patterns every session — wasting tokens and context. His solution is the "skill" primitive: a folder containing a `SKILL.md` file that packages instructions, context, and runnable scripts the agent loads on demand. Crucially, when the agent invokes the bundled Python CLI, only the script's output enters context — not the script's code itself — keeping execution efficient and repeatable.
He demonstrates it against the Open-Meteo weather API, running one command against the project's published `openapi.yml` to produce a ready-to-use skill in `./output/open-meteo/`.
To make skill creation fast, Jain built `api-skill-creator`, an open-source tool with a single dependency (`pyyaml`) that accepts OpenAPI 3.x specs, Swagger 2.0, Postman collections, or plain HTML documentation pages and outputs a complete `SKILL.md` plus a Python CLI. The tool works offline with no external API calls. He demonstrates it against the Open-Meteo weather API, running one command against the project's published `openapi.yml` to produce a ready-to-use skill in `./output/open-meteo/`. Dropping the output into `~/.claude/skills/` lets Claude Code immediately answer natural-language requests like fetching a 7-day weather forecast without any manual API translation from the developer.
Jain emphasizes that the Agent Skills format is an open standard, not a Claude-only construct. He lists Claude Code, Cursor, Codex CLI, OpenCode, Antigravity, and GitHub Copilot as compatible agents, meaning a skill generated once can be shared across an entire team whose members use different tools. He positions the practical shift as moving from "can I automate this?" to "what's worth automating?" — with operational API workflows (trigger a job, poll status, fetch a result, update a record) as the primary target.
Key facts
- 01A "skill" is a folder containing a `SKILL.md` file that packages API instructions, auth context, and an optional Python CLI script for a coding agent to load on demand.
- 02`api-skill-creator` generates a complete skill (SKILL.md + Python CLI) from OpenAPI 3.x, Swagger 2.0, Postman collections, or HTML docs pages with a single command and no external API calls.
- 03The tool has only one dependency (`pyyaml`) and works fully offline.