huma -- 一个端点都不漏的棘轮

huma – 一个端点都不漏的棘轮 Image: AI generated

如果你让AI Agent写API测试，它连一半都没做完就宣布"完成了"，如果你想让每一个端点都无一遗漏地被测试，如果你想用棘轮强制Hurl测试的编写——huma就是那个工具。

The Agent That Stops at Endpoint 15

Command	Role
`huma scan`	Auto-detects `openapi.yaml` and scans endpoints
`huma scan --from <file>`	Scans from a specified file
`huma next`	Shows next incomplete endpoint, or verifies and advances
`huma verify`	Runs current endpoint’s Hurl test; advances on pass
`huma status`	Shows progress (TODO/PASS/IMPROVE/DONE)
`huma prompt`	Outputs agent prompt for current TODO (no side effects)

openapi.yaml ──► huma scan ──► session
                                 │
                           huma next ◄──┐
                             │          │
                     ┌───────┴───────┐  │
                     │  TODO         │  │
                     │  (no .hurl)   │  │
                     └───────┬───────┘  │
                       agent writes    │
                       .hurl           │
                             │          │
                     ┌───────┴───────┐  │
                     │  PASS/IMPROVE │──┘
                     └───────────────┘

huma scan
# Scanned 42 endpoints

huma next
# TODO  GET /api/v1/users

huma status
# 15/42 PASS  |  0 IMPROVE  |  27 TODO

State	Meaning
TODO	No .hurl file
IMPROVE	Hurl exists but status codes missing
PASS	All expected status codes covered
DONE	Coverage stalled after retries

Language	Adapter	Analyzer	`backend.lang`
Go	`GoAdapter`	go/ast	`go`
Python	`PythonAdapter`	regex	`python`
Node.js	`NodeAdapter`	regex	`node`

Prefix	Domain
`M-`	manifest.yaml
`E-`	Endpoint input
`H-`	Hurl file
`S-`	Session state
`A-`	Adapter/server

Legacy codebase
    │
    ▼
codistill ──► openapi.yaml
    │
    ▼
huma ──► hurl/*.hurl
    │
    ▼
yongol ──► refactored code

go install github.com/park-jun-woo/huma@latest

npx skills add park-jun-woo/huma

Code: github.com/park-jun-woo/huma

参考文献

Jimenez, C. E. et al. (2024). SWE-bench. ICLR 2024. arxiv.org/abs/2310.06770
Xu, F. et al. (2024). TheAgentCompany. arxiv.org/abs/2412.14161
Lu, Q. et al. (2025). Runaway is Ashamed, But Helpful. arxiv.org/abs/2505.17616
Golmohammadi, A. et al. (2024). Testing RESTful APIs: A Survey. TOSEM. dl.acm.org/doi/10.1145/3617175
Martin-Lopez, A. et al. (2019). Test Coverage Criteria for RESTful Web APIs. dl.acm.org/doi/10.1145/3340433.3342822
Atlidakis, V. et al. (2019). RESTler. ICSE 2019. dl.acm.org/doi/10.1109/ICSE.2019.00083
Arcuri, A. (2019). EvoMaster. TOSEM. dl.acm.org/doi/10.1145/3293455
Karlsson, S. et al. (2020). QuickREST. ICST 2020. arxiv.org/abs/1912.09686
Corradini, D. et al. (2021). Restats. ICSME 2021. arxiv.org/abs/2108.08209
Scholten, T. et al. (2025). WuppieFuzz. arxiv.org/abs/2512.15554
Atlidakis, V. et al. (2021). REST API Fuzzing by Coverage Level. arxiv.org/abs/2112.15485

变更历史

2026-05-26: 初版

The Agent That Stops at Endpoint 15

参考文献

相关文章

变更历史