cenbonew/zing relay audit
Audit whether an LLM relay serves its claimed model (货不对板检测); fail CI when risk meets your threshold.
View on GitHubTrust Signals
- Scorecard Score
- not yet scored
- Maintenance Recency
- Activelast commit Jun 8, 2026
- License
- Apache 2.0
Pinned Snippet
uses: cenbonew/zing@fadd570cbea4d6af6180bbab97ae658430d1df30 # v0.11.0tags can be moved; commit SHAs can't. why a SHA?
Inputs
| name | description | required | default |
|---|---|---|---|
| base-url | Relay base URL, e.g. https://relay.example.com/v1. | yes | — |
| api-key | Relay API key. Pass a caller secret (e.g. ${{ secrets.RELAY_API_KEY }}). It is forwarded to zing via an environment variable and is never echoed. | yes | — |
| model | Model id actually sent in requests. | yes | — |
| claimed-model | Model the relay claims to serve, if different from `model`. | no | "" |
| api | Wire protocol: auto | openai | anthropic. | no | auto |
| declared-provider | Provider hint for KB lookup (openai, anthropic, deepseek, ...). | no | "" |
| suite | Detector suite: smoke | standard | deep | full. | no | standard |
| fail-on-risk | Fail the job when risk >= this level: low | medium | high. | no | high |
| python-version | Python version used to install and run zing. | no | 3.12 |
| version | pip install spec for zing. Defaults to the latest published `zing-audit`. Pin it for reproducible runs, e.g. `zing-audit==0.8.0`. | no | zing-audit |
Outputs
| name | description |
|---|---|
| risk | Risk level reported by zing (clean | low | medium | high | inconclusive). |
| score | Overall score 0-100 (may be empty if zing could not score the relay). |
| rating | Human-readable rating from the verdict (may be empty). |