Analyze Test Run

Analyze Test Run

Downloads artifacts from a GitHub Actions integration test run, generates a summarized skill invocation report, and files GitHub issues for each test failure with root-cause analysis.

When to Use

• Summarize results of a GitHub Actions integration test run
• Calculate skill invocation rates for the skill under test
• For azure-deploy tests: track the full deployment chain (azure-prepare → azure-validate → azure-deploy)
• Compare skill invocation across two runs
• File issues for test failures with root-cause context

Input

MCP Tools

All tools use owner: "microsoft" and repo: "GitHub-Copilot-for-Azure" as fixed parameters. method selects the operation within the tool.

Workflow

Phase 1 — Download & Parse

1. Extract the numeric run ID from the input (strip URL prefix if needed)

2. Fetch run metadata using the MCP actions_get tool:

   actions_get({ method: "get_workflow_run", owner: "microsoft", repo: "GitHub-Copilot-for-Azure", resource_id: "<run-id>" })

3. List artifacts using the MCP actions_list tool, then download each relevant artifact:

   // List artifacts
   actions_list({ method: "list_workflow_run_artifacts", owner: "microsoft", repo: "GitHub-Copilot-for-Azure", resource_id: "<run-id>" })
   // Download individual artifacts by ID
   actions_get({ method: "download_workflow_run_artifact", owner: "microsoft", repo: "GitHub-Copilot-for-Azure", resource_id: "<artifact-id>" })

   The download returns a temporary URL. Fetch the ZIP archive from that URL and extract it locally. If the environment restricts outbound HTTP (e.g. AWF sandbox), record in the analysis report that artifact content was unavailable and fall back to job logs via the get_job_logs MCP tool.

4. Locate these files in the downloaded artifacts:

   • junit.xml — test pass/fail/skip/error results
   • *-SKILL-REPORT.md — generated skill report with per-test details
   • agent-metadata-*.md files — raw agent session logs per test

   ⚠️ Note: If artifact ZIP files cannot be downloaded due to network restrictions, or if downloaded files cannot be extracted, use the get_job_logs MCP tool to identify test failures and produce a best-effort analysis from whatever data is accessible.

Phase 2 — Build Summary Report

Produce a markdown report with four sections. See report-format.md for the exact template.

Section 1 — Test Results Overview

Parse junit.xml to build:

Include a per-test table with name, duration (from time attribute, convert seconds to Xm Ys), and Pass/Fail result.

Section 2 — Skill Invocation Rate

Read the SKILL-REPORT.md "Per-Test Case Results" sections. For each executed test determine whether the skill under test was invoked.

The skills to track depend on which integration test suite the run belongs to:

azure-deploy integration tests — track the full deployment chain:

Build a per-test invocation matrix (Yes/No for each skill) and compute rates:

The azure-deploy integration tests exercise the full deployment workflow where the agent is expected to invoke azure-prepare, azure-validate, and azure-deploy in sequence. This three-skill chain tracking is specific to azure-deploy tests only.

All other integration tests — track only the skill under test:

For non-deploy tests (e.g. azure-prepare, azure-ai, azure-kusto), only track whether the primary skill under test was invoked. Do not include azure-prepare/azure-validate/azure-deploy chain columns.

Section 3 — Report Confidence & Pass Rate

Extract from SKILL-REPORT.md:

• Skill Invocation Success Rate (from the report's statistics section)
• Overall Test Pass Rate (from the report's statistics section)
• Average Confidence (from the report's statistics section)

Section 4 — Comparison (only when a second run is provided)

Repeat Phase 1–3 for the second run, then produce a side-by-side delta table. See report-format.md § Comparison.