• Home
  • Features
  • Pricing
  • Docs
  • Announcements
  • Sign In

kubeflow / mcp-server / 44542693061
95%

Build:
DEFAULT BRANCH: main
Ran 19 Nov 2026 02:21PM UTC
Jobs 3
Files 58
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst

19 Nov 2026 02:20PM UTC coverage: 94.742% (+0.02%) from 94.725%
44542693061

push

github

web-flow
feat(test): add Tier 1 safety judges for the confirm gate and personas (#346)

* feat(test): add Tier 1 safety judges for the confirm gate and personas

ARCHITECTURE.md lists rule-based safety judges as Tier 1 of the eval pipeline,
run on every PR. This adds them as plain pytest under tests/eval, with no
cluster, no LLM and no new dependency.

The per-tool unit tests check each preview one tool at a time. These judges
check the rules that hold across all tools, through an in-memory MCP client:

- a tool is read-only exactly when it has no confirmed parameter, and
  confirmed defaults to False
- confirmed=False returns a preview and sends no write request, for every
  mutating tool, with a control that confirmed=True does write
- each persona exposes exactly its allowlist, and hidden tools cannot be called
- the same checks through execute_tool in progressive and semantic modes
- every mutating tool must be listed in GATE_ARGS, so a new tool fails the
  suite until it is covered

Kubernetes is faked at ApiClient.call_api, which both the SDK and the raw
clients go through. No tool name, schema or response changes.

Signed-off-by: Abhishek <abhikokadwar2@gmail.com>

* fix(test): load every registered client in the safety judges

The judges built the server with the default client list, which is trainer
only. On the Spark branch (#51) they did not look at the Spark tools at all,
and the persona check failed because the policy lists Spark tools that a
trainer-only server does not expose.

Build every server with all of CLIENT_MODULES. A new client's mutating tools
are then checked without editing the test file, and the suite fails until
their arguments are added to GATE_ARGS.

Nothing changes on main: 41 checks pass, since the optimizer and hub stubs
register no tools.

Signed-off-by: Abhishek <abhikokadwar2@gmail.com>

* fix(test): judge hidden read-only tools in the persona checks

The hidden-tool checks only tried tools from GATE_ARGS, which holds mutat... (continued)

5352 of 5649 relevant lines covered (94.74%)

2.84 hits per line

Jobs
ID Job ID Ran Files Coverage
1 python-3.10 - 44542693061.1 19 Nov 2026 02:22PM UTC 58
94.74
GitHub Action Run
2 python-3.11 - 44542693061.2 19 Nov 2026 02:21PM UTC 58
94.74
GitHub Action Run
3 python-3.12 - 44542693061.3 19 Nov 2026 02:21PM UTC 58
94.74
GitHub Action Run
Source Files on build 44542693061
  • Tree
  • List 58
  • Changed 1
  • Source Changed 0
  • Coverage Changed 1
Coverage ∆ File Lines Relevant Covered Missed Hits/Line
  • Back to Repo
  • Github Actions Build #44542693061
  • 80beaea6 on github
  • Prev Build on main (#44467871744)
  • Next Build on main (#44549728917)
  • Delete
STATUS · Troubleshooting · Open an Issue · Sales · Support · CAREERS · ENTERPRISE · START FREE TRIAL · SCHEDULE DEMO
ANNOUNCEMENTS · TWITTER · TOS & SLA · Supported CI Services · What's a CI service? · Automated Testing

© 2026 Coveralls, Inc