chore(CI): Debug artifact download

[mtodor]

Description

Fixes the download-artifact path issue in the model-evaluation workflow and adds debug tooling. download-artifact@v8 extracts a single artifact directly into the target path without creating a per-artifact subdirectory, breaking the expected path eval-results/eval-results-<model>/.... Fixed by checking both possible file locations: the subdirectory path (multiple artifacts) and the flat path (single artifact).

Validation

• actionlint passes on the modified workflow file
• Verify artifact directory structure in CI logs - https://github.com/stackrox/stackrox-mcp/actions/runs/24250717708/job/70810127752

=== eval-results directory structure ===
eval-results:
mcpchecker-stackrox-mcp-e2e-out.json

AI-assisted development prompts

This PR was developed with AI assistance. Below are the prompts used:

Prompt: "This still does not work. I get: 'Warning: No results found for model gpt-5-mini'. But everything is downloaded properly. We need to test this somehow on a PR. Let's disable temporally creation of PR. List files and trigger it on PR. Research problem and prepare ticket to implement solution."
Response: Added temporary pull_request trigger, debug ls -R eval-results step, and commented out the PR creation step. Created beads task for the investigation.

Prompt: "This is what we got: eval-results: mcpchecker-stackrox-mcp-e2e-out.json"
Response: Confirmed that download-artifact@v8 places a single artifact flat at the target path with no subdirectory. Created fix ticket stackrox-mcp-ovo to handle both path layouts.

Prompt: "do it" (implement the fix)
Response: Applied the fix: checks subdirectory path first, falls back to flat path. Reverted all debug changes (removed pull_request trigger, debug step, restored PR creation).

Prompt: "Let's also put pattern matching back"
Response: Restored pattern: eval-results-* to the download-artifact step.

[codecov-commenter]

❌ 2 Tests Failed:

Tests completed	Failed	Passed	Skipped
361	2	359	12

View the full list of 2 ❄️ flaky test(s)

::policy 1

Flake rate in main: 100.00% (Passed 0 times, Failed 12 times)

Stack Traces | 0s run time

- test violation 1
- test violation 2
- test violation 3

::policy 4

Flake rate in main: 100.00% (Passed 0 times, Failed 12 times)

Stack Traces | 0s run time

- testing multiple alert violation messages 1
- testing multiple alert violation messages 2
- testing multiple alert violation messages 3

To view more test analytics, go to the Test Analytics Dashboard
_{📋 Got 3 mins? Take this short survey to help us improve Test Analytics.}

[github-actions]

E2E Test Results

Commit: 935f0a9
Workflow Run: View Details
Artifacts: Download test results & logs

=== Evaluation Summary ===

  ✓ list-clusters (assertions: 3/3)
  ✓ cve-detected-workloads (assertions: 3/3)
  ✓ cve-detected-clusters (assertions: 3/3)
  ✓ cve-nonexistent (assertions: 3/3)
  ✓ cve-cluster-does-exist (assertions: 3/3)
  ~ cve-cluster-does-not-exist (assertions: 2/3)
      - ToolsUsed: Required tool not called: server=stackrox-mcp, tool=, pattern=list_clusters
  ✓ cve-clusters-general (assertions: 3/3)
  ✓ cve-cluster-list (assertions: 3/3)
  ✓ cve-log4shell (assertions: 3/3)
  ✓ cve-multiple (assertions: 3/3)
  ✓ rhsa-not-supported (assertions: 2/2)

Tasks:      11/11 passed (100.00%)
Assertions: 31/32 passed (96.88%)
Tokens:     ~54953 (estimate - excludes system prompt & cache)
MCP schemas: ~12738 (included in token total)
Agent used tokens:
  Input:  18203 tokens
  Output: 21390 tokens
Judge used tokens:
  Input:  50449 tokens
  Output: 34291 tokens