Align the retagged generate-schema Flow eval before preview-catalog use

[tmatup]

Context

tests/tasks/uipath-maestro-flow/connector_features/generate_schema.yaml is tagged uipath-platform, so the 112-task Maestro Flow same-ground campaign correctly excludes it. The task should stay untouched in that campaign.

The 2026-08-20 preview-catalog run nevertheless shows two known alignment problems that its owner should resolve before using it in a cross-skill comparison:

1. Its artifact check requires **/GenerateSchemaTest*.flow, so it cannot discover the preview Flow SDK's valid root-level emit-only layout.
2. Two gating criteria require live-v1 CLI commands (uip is resources describe and uip maestro flow validate) instead of the persisted GenerateSchema outcome.

Evidence: runs/adhoc-flow-v2-preview-20260820/default/skill-flow-ipe-generate-schema/00/task.json recorded score 0.0, no discovered Flow, and neither command-telemetry gate.

Requested follow-up

Apply the same neutral measurement recipe used by the Maestro Flow expansion:

• discover both solution-wrapped and root-level Flow artifacts;
• grade the persisted Jira GenerateSchema structure and validation outcome;
• keep loop-specific command telemetry report-only when no durable equivalent exists;
• preserve criterion intent and weights.

This issue is an owner note only. The current Maestro Flow campaign does not edit or retag the task.

🤖 Generated with Claude Code
Co-Authored-By: Claude

[tmatup]

Implemented at the enabling layer in #2756, commit 6d2b9cc10.

The fix:

• routes GenerateSchema grading through the shared solution-or-root Flow discovery contract;
• parses customFieldsRequestDetails from the Jira node's persisted configuration, so unrelated raw tokens cannot spoof the result;
• makes the live-v1-only is resources describe command telemetry advisory;
• replaces command-history validation with an independent validation of the persisted artifact;
• adds positive coverage for root SDK emits and solution-wrapped flows plus negative controls for missing parent values and misplaced tokens.

Verification is green: all 287 Maestro Flow checker tests pass, coder-eval accepts the revised task schema, the CLI verb audit is clean, and a recorded live-v1 GenerateSchema artifact passes the strengthened checker. The PR now closes this issue on merge; no overlapping second PR is needed.

🤖 Generated with Claude Code
Co-Authored-By: Claude

[tmatup]

The complete enabling fix is ready in #2756. In addition to the structural GenerateSchema checker correction, commit b3544e685 fixes shared-helper imports for isolated coder-eval task mounts and adds a corpus-wide regression guard.

Current-head evidence: all 289 Maestro Flow checker tests pass, the task schema validates, and every reported CI check is successful or intentionally skipped. The PR is awaiting required human approval and will close this issue on merge.

🤖 Generated with Claude Code
Co-Authored-By: Claude