test-fix-workflow

name	test-fix-workflow
description	Workflow for diagnosing, fixing, and verifying failing Terraform acceptance tests from GitHub issue URLs (detecting test-failure labels), direct prompts, or log files using Failure Scenario Decision Trees.

test-fix-workflow

This document guides the agent through the structured 3-step lifecycle for resolving Terraform acceptance test failures in Magic Modules using scenario classification.

Consult .agents/knowledge/index.md for the topics the failure touches and open the relevant sources.

---

Essential Reference Documentation

When diagnosing and fixing test failures, consult these core documentation guides in docs/content/ (each contains tabbed examples for MMv1 Generated vs Handwritten code):

• docs/content/develop/diffs.md — Authoritative guide for fixing plan diffs, permadiffs, API server defaults, and ignore_read fields (Scenario 2).
• docs/content/test/test.md — Authoritative guide for acceptance test conventions, dynamic randomization, dependency bootstrapping, and VCR skipping rules (Scenario 5).
• docs/content/test/run-tests.md — Authoritative guide for compiling downstream providers and running make testacc verification loops.

---

Execution Steps

1. Failure Information Intake

• Execute the intake-test-failure skill (.agents/skills/utils/intake-test-failure/SKILL.md) on the input provided by the user (GitHub issue URL, direct prompt text, GCS log link, or local log file).
• Check for GitHub issue labels such as test-failure, test-failure-100, test-failure-50, or any test-failure-* label (e.g., test-failure-0, test-failure-10) to confirm test failure classification.
• Inspect issue failure rates to determine target_provider (ga, beta, or both).
• Produce the Normalized Failure Payload:

  normalized_failure_payload:
    test_name: "<ExactTestFunctionName>"
    target_provider: "ga" # "ga", "beta", or "both"
    error_message: |
      <Full error output, go test backtrace, and stdout plan diff for GA and/or Beta>
    parsed_logs_dir: "debug_output/<test_name>/" # (Optional)
  

---

2. Failure Scenario Classification & Remediation (Choose Path)

Match failure symptoms against the central decision tree catalog in .agents/skills/utils/test-failure-decision-tree/SKILL.md (all catalog scenarios).

Consult .agents/skills/utils/test-failure-decision-tree/SKILL.md for full symptom patterns, root causes, and remediation recipes.

Path A: Automated Subagent (Mandatory Default)

• Action: Invoke the test-fixer subagent (.agents/agents/test-fixer/) using the invoke_subagent tool.
• Prompt: Pass the Normalized Failure Payload to test-fixer.
• Wait: The subagent will classify the failure scenario, consult .agents/knowledge/index.md for relevant design rules, perform remediation (including automatically running gcloud services enable for shared CI projects or editing magic-modules test configs for test-created secondary projects), run make provider VERSION=<ga|beta> and make build, and execute target acceptance tests for ga, beta, or both to verify PASS.
• Handoff:
  • If test-fixer reports success, present the fix summary to the user.
  • If test-fixer reports unresolved issues, switch to Path B (Interactive Debugging) and consult .agents/skills/utils/test-failure-decision-tree/SKILL.md.

Path B: Interactive Debugging (Fallback)

• Use the qa-test-runner subagent to isolate logs and inspect request/response JSONs.
• Match symptoms against the decision tree catalog in .agents/skills/utils/test-failure-decision-tree/SKILL.md (all catalog scenarios) and apply source modifications using triage (.agents/skills/operations/triage/) and fix (.agents/skills/operations/fix/) skills.
• Execute provider generation (make provider VERSION=<ga|beta>) and re-run acceptance tests for each target provider version (ga, beta, or both) to verify PASS.

---

3. Verification & Compliance Checks

• Verify that test output reports PASS for all failing targets (ga, beta, or both).
• Verify Handwritten Test Naming Compliance: For handwritten test files (*_test.go), verify that any newly added unit tests are named Test... (without the TestAcc prefix) and handwritten acceptance tests are named TestAcc.... (Generated acceptance tests from MMv1 samples handle naming automatically).
• Verify Remediation Scope: Verify that code modifications are strictly scoped to the specific field(s) or resource(s) proven by the test failure log or plan diff to be causing the failure. Do NOT include assumption-based edits to adjacent/sibling fields without empirical evidence.
• Verify Breaking Change Compliance & User Notification: Consult all files in docs/content/breaking-changes/ (breaking-changes.md and make-a-breaking-change.md) before making any schema or behavioral modifications. If a breaking change needs to be made, explicitly state it to the user and list out the reason why, referencing the applicable policy in docs/content/breaking-changes/.
• Verify that no test-dodging flags (ignore_read, default_from_api, etc.) were introduced without justification.