ai-engineering-fluency: Agent for Claude Code

Improve Mutation Score Agent

Analyzes the latest Stryker mutation report and adds targeted tests to kill surviving mutants — but only where the test adds real value, not to chase a score.

---

Step 1 — Download the latest mutation report artifact

Use the GitHub CLI to find and download the most recent mutation-report artifact from CI:

# List recent runs that produced a mutation-report artifact
gh run list --workflow=ci.yml --limit=10 --json databaseId,displayTitle,conclusion,createdAt

# Download the artifact from the most recent successful run
gh run download <run-id> --name mutation-report --dir /tmp/mutation-report

The artifact contains:

• mutation.json — full machine-readable report (this is what we analyze)
• report.html — human-readable HTML report

If the CI artifact is unavailable, run Stryker locally instead:

cd vscode-extension
pnpm run compile-tests
pnpm exec stryker run
# Report is written to: vscode-extension/reports/mutation/mutation.json

---

Step 2 — Parse the mutation report

Read /tmp/mutation-report/mutation.json (or vscode-extension/reports/mutation/mutation.json).

The structure is:

{
  "mutationScore": null,   // may be null — compute from counts instead
  "files": {
    "out/src/tokenEstimation.js": {
      "mutants": [
        {
          "id": "1",
          "mutatorName": "StringLiteral",
          "replacement": "\"\"",
          "location": { "start": { "line": 12, "column": 4 }, "end": { "line": 12, "column": 20 } },
          "status": "Survived",   // Survived | Killed | Timeout | NoCoverage
          "description": "Replaced \"some string\" with \"\""
        }
      ]
    }
  }
}

Extract all mutants with "status": "Survived", grouped by source file. Compute the score as: killed / (killed + survived + timedOut) * 100

---

Step 3 — Map surviving mutants back to TypeScript source

The report references compiled JS files (e.g. out/src/tokenEstimation.js). Map these to their TypeScript equivalents: