Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions agentops-cli/src/commands/dashboard.js
Original file line number Diff line number Diff line change
Expand Up @@ -557,6 +557,7 @@ function validateDashboardUx() {
policy_review: true,
semantic_review: true,
promotion_approvals: true,
robust_eval_center: true,
alert_handoff_review: true,
schema_version_coverage: true,
exporter_failure_review: true,
Expand Down
34 changes: 34 additions & 0 deletions agentops-cli/src/commands/product.js
Original file line number Diff line number Diff line change
Expand Up @@ -228,6 +228,40 @@ function productAudit(options = {}) {
ux.errors
));

checks.push(check(
'robust-eval-center-contract',
ux.ok
&& ux.contracts?.robust_eval_center === true
&& fileIncludes('agentops-cli/src/legacy.js', [
'hiddenCheckPacks',
'permissionProfiles',
'macos-network-blocked',
'commandFileSeal',
'semanticChecks',
'llm-judge',
'artifactDiff',
'promotionApproval'
])
&& fileIncludes('grafana/dashboards/v2/08-evals-quality.json', [
'BenchmarkArtifactContentDiffs',
'BenchmarkApproval'
])
&& fileIncludes('docs/agentops-architecture-product-audit.md', [
'Robust Eval Center',
'Hidden check packs',
'Rubric and semantic scoring',
'Artifact diffing',
'Promotion policy',
'Cross-platform managed OS-level'
]),
[
'agentops-cli/src/legacy.js',
'grafana/dashboards/v2/08-evals-quality.json',
'docs/agentops-architecture-product-audit.md'
],
ux.errors
));

checks.push(check(
'azure-ingest-privacy-plan',
fileIncludes('agentops-cli/src/lib/azure/v2-ingest-plan.js', ['--allow-content', 'AgentOpsContent_CL', 'schema_versioning', 'schema_migration_policy'])
Expand Down
1 change: 1 addition & 0 deletions agentops-cli/test/index.test.js
Original file line number Diff line number Diff line change
Expand Up @@ -4332,6 +4332,7 @@ test('product audit proves the local AgentOps control-room contract', () => {
'grafana-v2-pack',
'kql-library',
'run-centric-ui-contract',
'robust-eval-center-contract',
'content-transcript-opt-in',
'first-run-loop',
'ask-agentops-response-flow',
Expand Down
20 changes: 13 additions & 7 deletions docs/agentops-architecture-product-audit.md
Original file line number Diff line number Diff line change
Expand Up @@ -1239,15 +1239,21 @@ Success criterion:
Users can compare baseline vs candidate agents and detect safety, quality, cost, and anti-cheat regressions before promotion.
```

Implemented:

- Hidden check packs are supported as masked, sealed check packs, with signed fixture pack manifests, trust roots, revoked signing key IDs, and rotation windows.
- Rubric and semantic scoring supports deterministic file-content, regex, file-rubric, and command-backed `llm-judge` checks with reusable hosted judge provider command templates.
- Network and permission controls include per-task permission profiles, read-only copied fixtures, allowed-tool risk policies, and opt-in macOS `sandbox-exec` network blocking that fails closed when requested on unsupported hosts.
- Local harness tamper resistance includes copied fixture workspaces, read-only benchmark profiles, sealed fixture packs, and `commandFileSeal` checks for test scripts and command files.
- Artifact diffing is available in CLI reports and the Evals & Quality dashboard, including per-file review and capped content diff previews.
- Scorecard UI is present in the Evals & Quality dashboard through eval scorecards, regression follow-up, before/after comparisons, hidden check review, policy review, semantic review, artifact review, and promotion approvals.
- Promotion policy can require gate results, approval evidence, approver identity/counts, approved external review metadata, and optional external review verification for GitHub PRs, Azure DevOps PRs, and Jira issues.
- `agentops product audit` now includes `robust-eval-center-contract` so this eval-center shape remains a local release gate.

Required work:

- Hidden tests.
- Rubric/semantic scoring beyond deterministic file-content adapters.
- Network and permission profiles.
- Managed immutable harness isolation beyond command-file seals.
- Artifact diffing.
- Scorecard UI.
- Promotion policy.
- Cross-platform managed OS-level immutable harness and network egress isolation beyond the current copied fixture, read-only profile, command-file seals, signed fixture packs, and opt-in macOS network sandbox.
- Hosted judge service deployment for `llm-judge` beyond the current non-mutating Azure Container Apps provisioning plan.

### 5. Trustworthy Data Quality

Expand Down
Loading