Skip to main content
skill CX Codex

volcengine-sdk-generator — Claude Code skill: what it does

The volcengine-sdk-generator Claude Code skill finds Volcengine APIs and generates SDK call examples. It activates for code requests in Python, Go, Java, PHP, cURL, or Node.js, and for SDK configuration questions.

MIT 112 installs ★ 37 updated 2026-11-16

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add volcengine/volcengine-agent-toolkit --skill volcengine-sdk-generator -a claude-code
Open on GitHub

What volcengine-sdk-generator does

The skill searches API Explorer, retrieves the selected API’s Swagger definition, then sends user-provided parameters to the code generator. The resulting example covers authentication setup, request parameters, the API call, and printing its response. Language-specific reference files cover questions about retries, timeouts, proxies, SSL, and other SDK settings.

API discovery runs online only. If search fails, the skill reports the error rather than choosing an API from a bundled catalog. Request parameters come from the user; when required fields are missing, it can insert mock values and mark them as such. Questions limited to API parameters, response schemas, or API comparisons go to volcengine-api; CLI operations go to volcengine-cli.

What it helps with

  • find a Volcengine API from its name or action
  • generate a Python or Go API call example
  • create SDK samples for Java, PHP, or Node.js
  • configure SDK retries, timeouts, or request signing

Example prompts for the agent

  • Generate a Python example for the Volcengine AssumeRole API using these parameters: …
  • Write a Go call for this Volcengine API action: …
  • How do I configure timeouts and retries in the Volcengine Python SDK?

How to install volcengine-sdk-generator

1Skills CLI (Vercel)

npx skills add volcengine/volcengine-agent-toolkit --skill volcengine-sdk-generator -a claude-code

2Manually: put the files in place

~/.claude/skills/volcengine-sdk-generator/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/volcengine-sdk-generator/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about volcengine-sdk-generator

How do I install volcengine-sdk-generator in Claude Code?

Run npx skills add volcengine/volcengine-agent-toolkit --skill volcengine-sdk-generator -a claude-code. The skill file is at ~/.claude/skills/volcengine-sdk-generator/SKILL.md.

What does volcengine-sdk-generator do?

It locates a Volcengine API through API Explorer, fetches its Swagger definition, and generates an SDK call example from user-supplied parameters. Supported languages include Python, Go, Java, PHP, cURL, and Node.js.

Which requests does this skill hand off to other skills?

Requests limited to API parameters, error codes, response schemas, or API comparisons go to volcengine-api. CLI-based operations go to volcengine-cli.

Source text

SKILL.md · MIT
---
name: volcengine-sdk-generator
description: >
  Generate accurate Volcengine SDK examples by locating an API through API Explorer search,
  fetching its swagger, and calling api/common/explorer/make-code with user-provided
  Params. Use when the user asks how to write Volcengine SDK code, generate SDK samples, or call a
  Volcengine API in Python, Go, Java, PHP, cURL, Node.js. Supports Chinese and
  English API names such as "角色扮演", "AssumeRole", or "STS AssumeRole". Also use for explicit SDK
  configuration questions about retry, timeout, AK/SK, STS, AssumeRole, temporary credentials,
  proxy, connection pooling, SSL, debug mode, request signing, response parsing, and error handling.
  If the user only needs API parameters, enum values, required fields, error codes, response
  schemas, pagination, or API comparisons, hand off to volcengine-api. If they need CLI-based
  operations, hand off to volcengine-cli.
license: MIT
---

# Volcengine SDK Generator

Generate ordinary Volcengine API call examples by default. Keep the default output focused on the
API call itself: authentication setup, request object/params, invocation, and printing the response.
Do not add retry, proxy, connection pool, debug logging, or other advanced SDK configuration unless
the user explicitly asks for those topics.

For advanced SDK configuration questions, use the language-specific files under `references/`.
These reference files are self-contained; answer from them directly.

| Topic | Reference |
| --- | --- |
| Go SDK configuration | `references/sdk-integration-go.md` |
| Python SDK configuration | `references/sdk-integration-python.md` |
| Java SDK configuration | `references/sdk-integration-java.md` |
| Node.js SDK configuration | `references/sdk-integration-nodejs.md` |
| PHP SDK configuration | `references/sdk-integration-php.md` |

## Core Rules

- API discovery uses `scripts/rg_rank.py`, which calls API Explorer search with both `Channel=service` and `Channel=api`. After a concrete API is selected, call `scripts/make_code.py` in direct mode with `--service-code`, `--api-version`, and `--action`.
- API discovery is online-only. The search endpoint does not require cookies, login state, UID, DID, UUID, or browser-specific headers. The script sends only `Query`, `Channel`, `Limit`, and `Offset` as query parameters.
- `rg_rank.py` preserves `Channel=api` order except for version selection: when the same `service_code + action` has multiple versions and the user did not specify one, it moves the result with `IsRecommended=1` ahead of the other versions in that group. It also includes `Channel=service` results as separate service metadata.
- If online search fails, report the returned search errors. Do not guess an API or fall back to a bundled catalog.
- Use the user's original request as `--query`; do not rewrite it into a guessed API name or SDK method.
- Extract a shorter `--query` only when the user clearly self-corrects with words such as `等等`, `不对`, `重新来`, `我改主意`, `其实想要`, `其实是想要`, `应该是`, or `改成`; the extracted query must be a continuous substring from the user request, not a guessed API name.
- When multiple resources or services appear without a clear self-correction, run `rg_rank.py` once with the original request. If the returned candidates remain ambiguous, list them concisely and ask the user to choose.
- Pure-generic queries such as `api`, `sdk`, or `demo` are sent online but are likely to be ambiguous. Refine the query with a concrete resource or action word.
- ServiceCode is case-sensitive. Preserve the exact `service_code` value from the selected API record when calling `make_code.py`; for example, `Kafka` is not `kafka`.
- When the same `service_code + action` appears in multiple API versions and the user did not specify a version, select the result with `is_recommended=true` even if its `remote_rank` is lower. If the user specified a version, use that exact version.
- Fetch `api-swagger` only after a concrete API is selected.
- The `api-swagger` fetch may include both `Version` and `APIVersion` because that endpoint expects them; the `make-code` payload must not include `Version`.
- Call `make-code` with `ApiAction`, `ServiceCode`, `APIVersion`, `Region`, and `Params`.
- Do not send `Version` to `make-code`.
- `Params` must come from the user. Do not auto-fill from swagger demos.
- Swagger is used for API metadata, required parameter hints, and lightweight validation. Do not serialize query arrays, form arrays, or `.N` parameters for `make-code`; pass the user’s JSON object as `Params`.
- If the user omits required top-level parameters, mock only those top-level required parameters and clearly mark them as mock values. Do not mock optional parameters. If a required top-level parameter is an object or array, recursively fill only required child fields/items.
- Mock values must be derived from the fetched swagger first: prefer `example`, then `examples`, then `default`, then `enum`, then type/constraint-aware fallback. If a swagger example is masked, such as `****` or `XX`, skip it and use a valid fallback value.
- Mock comments must be in Chinese by default and placed near the mocked assignment line when the target language supports line comments; do not add a duplicate mock banner at the top of the returned code.
- Returned SDK code must print the API response value. The script post-processes fixed `make-code` templates for Python, Go, Java, and PHP to assign the response and print it.
- Return the code generated by `scripts/make_code.py` as the primary SDK example. Do not rewrite SDK request construction, authentication setup, or response handling into a custom application unless the user explicitly asks for that. If optional formatting or convenience logic is added, keep the generated active response print, such as `print(resp)`, `fmt.Println(resp)`, `System.out.println(resp)`, `print_r($response)`, or `console.log(response)`.
- If language is not specified, return all languages from `DemoSdk`.
- SDK install/dependency info is a separate, on-demand capability. Only when the user explicitly asks how to install the SDK or which package/version to use (keywords such as 依赖, 安装, install, 版本, 包, maven, `go get`, pip, composer, npm), run `scripts/sdk_info.py` with `--service-code` and `--api-version` (reuse the selected API's values; add `--language` to filter to one of go/python/java/php/nodejs). It returns the live install command (`RunCommand`) plus package and version per language. Do NOT run it during normal code generation.
- If the API match is ambiguous, show concise candidates and ask the user to choose one. Do not add follow-up execution boilerplate such as saying that you will fetch swagger, mock required params, and call `make-code` after the user chooses.

## Workflow

1. Parse the user request:
   - API hint: action, Chinese name, service code, or natural-language description.
   - Language: Python, Go, Java, PHP, cURL, Node.js, or unspecified.
   - Params: JSON object supplied by the user.
   - Region: user value or default `cn-beijing`.

2. Locate or validate the API once for every request. If the user already provides `service_code`, `api_version`, and `action`, use those exact values in direct mode; `make_code.py` verifies them through the online `api-swagger` response. For natural-language requests, call `scripts/rg_rank.py` with `--query` first. Use the full original request unless a clear self-correction keyword points to a later target; in that case use the final target substring.

3. After selecting `service_code`, `api_version`, and `action`, call `scripts/make_code.py` in direct API mode with `--service-code`, `--api-version`, and `--action`. If top-level required params are missing, the script fills only those required top-level params with swagger-derived mock values and annotates the returned code in Chinese. Ask for clarification only when API selection is genuinely ambiguous in the online results; in that clarification, only list candidates and ask the user to choose.

4. Once params are available, call `make-code` through the script and return the official code. Do not transform the generated sample into a hand-written utility; any custom formatting must be secondary and must preserve the active response print.

## Commands

Search online candidates:

```bash
python3 scripts/rg_rank.py \
  --query '角色扮演' \
  --limit 10 \
  --format text
```

Direct API mode:

In TRN examples, `<account-id>` is the Volcengine account ID segment, such as a masked `2134xxxyyy`.

```bash
python3 scripts/make_code.py \
  --service-code sts \
  --api-version 2018-01-01 \
  --action AssumeRole \
  --language python \
  --params-json '{"DurationSeconds":3600,"RoleSessionName":"demo","RoleTrn":"trn:iam::<account-id>:role/demo"}'
```

Use `--refresh-swagger` when API Explorer metadata has just changed or generated code looks stale:

```bash
python3 scripts/make_code.py \
  --service-code sts \
  --api-version 2018-01-01 \
  --action AssumeRole \
  --refresh-swagger \
  --params-json '{}'
```

Fetch SDK install/dependency info on demand (only when the user asks about install, package, or version):

```bash
python3 scripts/sdk_info.py \
  --service-code sts \
  --api-version 2018-01-01 \
  --language go
```

© volcengine · MIT · volcengine/volcengine-agent-toolkit

Similar skills

All skills for Backend and APIs →

From the same set9

volcengine-cli volcengine-cli — a skill for Claude Code and Codex that automates Volcengine cloud resources via the ve command-line tool. It provisions ECS instances, handles VPC networks, and queries Ark models. CX Codex +1 1,300installs volcengine-find-skills The volcengine-find-skills Claude Code skill searches the Volcengine catalog and installs selected skills. It activates when a task needs a matching capability or the current Volcengine setup falls short. CX Codex +1 292installs volcengine-feedback The volcengine-feedback Claude Code skill drafts feedback about Volcengine cloud workflows and can submit it to GitHub with authorization. It can offer a feedback prompt after a CLI error or an unfinished task while the main work continues. CX Codex +1 216installs volcengine-api This Claude Code skill answers questions about Volcengine API specifications. It activates for requests about parameters, error codes, request methods, responses, or API differences. CX Codex +1 154installs volcengine-deploy The volcengine-deploy Claude Code skill deploys a local project or Git repository to Volcengine as a cloud service. It activates when asked to publish code, run an app in the cloud, or deploy to ECS, VKE, or veFaaS. CX Codex +1 122installs volcengine-prepare The volcengine-prepare Claude Code skill examines a project before deployment to Volcengine. It ranks viable options across ECS, VKE, and veFaaS. CX Codex +1 119installs volcengine-tls This Claude Code skill guides agents through Volcengine TLS tasks using volclog. It activates for log searches and analysis, plus work with log projects and topics. CX Codex +1 118installs volcengine-iac The volcengine-iac Claude Code skill builds Terraform configuration for Volcengine infrastructure. It activates when Terraform is chosen or a Terraform state already exists. CX Codex +1 111installs

Project sponsors

Companies that keep this analytics open