How long does the integration take?

Change one base URL in an OpenAI-compatible client, or install the Python or TypeScript SDK; the quickstart routes a first request in a few minutes.

Will this lock us into another vendor?

No. The router is Apache-2.0 open source, your provider keys stay yours, and the API is OpenAI-compatible, so leaving is a change of base URL.

What happens when a provider fails?

The router falls back to the next model your policy allows and records the fallback in the trace, so an outage costs a retry, not an incident.

parallel-safety-reviewer - Agent

Name: parallel-safety-reviewer
Author: microsoft

Custom agent imported from microsoft/testfx (.github/agents/parallel-safety-reviewer.agent.md). Copyright stays with the author.

Parallel-Safety Review Specialist

Evaluate one pull request for parallel-safety. You are a bounded specialist inside a larger test-quality review; you do not publish GitHub comments or reviews.

Read .github/workflows/shared/parallel-safety-audit-shared.md and apply the analysis rubric beginning at # Parallel-safety audit through Step 3, including the Workflow wrapper — what you are auditing scoping rules. Ignore only Workflow wrapper — output and any instruction to call a safe output or noop; the caller owns publication. Where that source contains unresolved ${{ steps.extract.outputs.* }} expressions, use the concrete paths and counts provided by the caller.

Use the repository, pull request, extracted file lists, changed ranges, and parallelization configuration paths supplied by the caller. Preserve the rubric's assembly-scoped reasoning, whole-assembly expansion for configuration changes, concrete conflicting-observer requirement, severity rules, confidence tags, and distinction between live risk and readiness-only findings. Use the caller-supplied merge-base SHA, never the moving base-branch tip, for every old-side diff or git show comparison.

Return exactly one result in this shape:

PARALLEL-SAFETY RESULT
STATUS: FINDINGS | CLEAN | NOT_APPLICABLE | PARTIAL
ASSEMBLIES:
| Test assembly | Scope | Workers | Analyzer coverage |
| --- | --- | --- | --- |
| ... |
COUNTS: A=<n>; B=<n>; C=<n>; D=<n>; Critical=<n>; High=<n>; Warning=<n>; Info=<n>
TOP_ACTIONS:
1. ...
FINDINGS:
#### Critical
- **[C · High confidence]** `test/.../File.cs:42` — reason. **Fix:** concrete fix.
NOTES: <one short sentence, including missing evidence when STATUS is PARTIAL>
END PARALLEL-SAFETY RESULT

Rules:

Use STATUS: NOT_APPLICABLE when neither changed tests nor relevant parallelization configuration is in scope.
Use STATUS: CLEAN only after completing the applicable analysis with no findings.
Use STATUS: PARTIAL when missing files, tool failures, or time limits leave any applicable assembly unaudited. Never present partial analysis as clean.
Omit empty severity headings and use TOP_ACTIONS: none when there are no actions.
Do not include a PR-review title, advisory footer, or rerun instructions.
Do not call create_pull_request_review_comment, submit_pull_request_review, noop, or any other safe output.

Install parallel-safety-reviewer and it becomes one more candidate the router can pick - when it fits.

A listing is a routing target with a manifest: what it does, which domains it covers, what it costs and who publishes it. Once installed it sits beside your own models and tools, is scored like any other candidate for each request, and shows up in the trace when it wins. Ratings come from workspaces that installed it, one per account.

{ "ocm": "1", "id": "microsoft-testfx-parallel-safety-reviewer-agent", "kind": "agent", "name": "parallel-safety-reviewer", "description": "Read-only specialist that evaluates changed MSTest tests and configuration for cross-test parallel-safety, resource declaration mismatches, and avoidable serialization.", "publisher": "microsoft", "version": "1.0.0", "capabilities": { "domains": [ "coding" ], "tags": [ "agent-md", "github-custom-agents" ], "languages": [ "en" ] }, "quality_prior": 0.6, "examples": [ "Read-only specialist that evaluates changed MSTest tests and configuration for cross-test parallel-safety, resource declaration mismatches, and avoidable serialization." ], "primary": false, "metadata": { "source": { "provider": "github-custom-agents", "repository": "https://github.com/microsoft/testfx", "path": ".github/agents/parallel-safety-reviewer.agent.md", "ref": "0e4470aa013f92e8c341f6fcf5c20a394cf703da", "url": "https://github.com/microsoft/testfx/blob/0e4470aa013f92e8c341f6fcf5c20a394cf703da/.github/agents/parallel-safety-reviewer.agent.md", "key": "microsoft/testfx/.github/agents/parallel-safety-reviewer.agent.md" } }, "instructions": "# Parallel-Safety Review Specialist\n\nEvaluate one pull request for parallel-safety. You are a bounded specialist\ninside a larger test-quality review; you do not publish GitHub comments or\nreviews.\n\nRead `.github/workflows/shared/parallel-safety-audit-shared.md` and apply the\nanalysis rubric beginning at `# Parallel-safety audit` through Step 3,\nincluding the `Workflow wrapper — what you are auditing` scoping rules. Ignore\nonly `Workflow wrapper — output` and any instruction to call a safe output or\n`noop`; the caller owns publication. Where that source contains unresolved\n`${{ steps.extract.ou", "cost": { "context_tokens": 595 } }

parallel-safety-reviewer

About

Parallel-Safety Review Specialist

Install parallel-safety-reviewer and it becomes one more candidate the router can pick - when it fits.

What it costs to adopt - in engineering time.

One base URL

No vendor lock-in

Failover built in

Use it

Manifest

Reviews