Use Fetcher Skills
Skills give a coding agent package boundaries, implementation workflows, and API references. This documentation helps you understand and review the result; a skill helps the agent perform the work within those boundaries.
Install the plugin
Fetcher’s skills/plugins.json publishes these skills as ahoo-fetcher-skills. Follow the Ahoo Skills installation guide for the matching plugin.
codex plugin marketplace add Ahoo-Wang/skills --ref main
codex plugin add ahoo-fetcher-skills@ahoo-skillsClaude Code uses /plugin marketplace add https://github.com/Ahoo-Wang/skills followed by /plugin install ahoo-fetcher-skills. Distribution controls the installed version; do not assume it immediately matches repository HEAD.
Choose the task boundary
| Task | Skill | API reference |
|---|---|---|
| HTTP clients | $fetcher-integration | fetcher |
| Declarative services | $fetcher-decorator-service | decorator |
| Event delivery | $fetcher-eventbus | eventbus |
| Stored values | $fetcher-storage | storage |
| SSE consumption | $fetcher-llm-streaming | eventstream |
| Chat completions | $fetcher-openai-client | openai |
| OpenAPI types | $fetcher-openapi-types | openapi |
| Client generation | $fetcher-openapi-generator | generator |
| React request state | $fetcher-react-hooks | react |
| Data viewers | $fetcher-viewer-components | viewer |
| CoSec authentication | $fetcher-cosec-auth | cosec |
| Wow commands and queries | $fetcher-wow-cqrs | wow |
Ask for a verifiable outcome
$fetcher-openapi-generator generate from ./openapi.yaml into src/generated.
Use the existing tsconfig, preserve the input contract, and type-check the output.Provide inputs, output location, existing clients, runtime, and acceptance criteria. A skill’s references/api.md supplies detailed material on demand; human Reference pages also provide accurate contracts without requiring readers to switch to a skill to develop an application.
Author source files under this repository’s skills/. Ahoo Skills generates distribution copies; change and validate the source rather than editing generated copies.
Start with runnable examples and task guides to give the agent concrete inputs and expected results. Use architecture to review boundaries.