QA Investigation
A persistent, file-backed investigation journal for a specific failing test. This is the execution layer: it resolves a concrete failure. It does not validate strategy or architecture (grill-me-qa) nor generate QA deliverables (qa-manual-istqb).
The core idea: your context window is volatile RAM; the filesystem is persistent disk. Writing goals, evidence, and decisions to markdown prevents context drift during a long investigation.
When to Use This Skill
- A test fails intermittently (flaky) or deterministically (bug), and you need the root cause.
- The investigation spans many tool calls, multiple runs, or more than one session.
- You want a durable record of what you found, decided, and why.
When NOT to Use This Skill
- Authoring a test from scratch — use the relevant automation/framework skill.
- Designing a framework or coverage strategy — strategy validation (
grill-me-qa) or artifact generation (qa-manual-istqb). - Simple questions or quick lookups (fewer than ~5 tool calls).
- General review of non-test production code.
The boundary is not "is it a selector / browser issue / timeout" — any of those can be worth investigating. The boundary is whether the request needs a persistent, multi-step root-cause investigation or is a one-shot tactical task. If uncovering the why takes evidence, runs, and iteration, use this skill.
Tool Agnosticism
This method is independent of any test framework — web, API, mobile, embedded, unit, load. Terms like "browser", "selector", "network requests", or "CI vs local" are illustrative, not requirements; substitute the equivalent in your stack.
Core Process
The phases are the same whether the failure is flaky or a deterministic bug. The skill discovers the classification during triage — it does not assume it up front.
Phase 1: Reproduction & Triage
- Reproduce reliably; isolate variables (parallelism, repeat count, environment, data/state).
- Determine: intermittent (flaky), deterministic (bug), or non-reproducible? This is a finding, not an input.
- Record the classification and the evidence that supports it.
- Goal: a confirmed reproduction or a documented non-reproducible failure.
Non-reproducible path: if the failure cannot be reproduced after a bounded number of attempts, do not force a label. Record it as non-reproducible with partial evidence, note the suspected nature (infrastructure, app logic, or test-side timing), and escalate or flag for observation. Log the decision and reason to
qa_investigation_findings.md. See Flow for detail.
Phase 2: Evidence Collection
- Capture logs, stack traces, screenshots, traces, retry counts, dependency activity, timings.
- Multimodal content (images, page/dependency data, PDFs) does not persist in context — write it to
qa_investigation_findings.mdas text immediately. - Redact sensitive data (tokens, cookies, credentials, email addresses, PII) before persisting; do not write raw screenshots, traces, logs, or network captures verbatim — summarize them in text with sensitive parts masked.
- Note environment specifics: build/version, platform, device, data conditions, worker count.
- Goal: enough evidence for a defensible hypothesis.
Phase 3: Hypothesis & Root Cause
- Form the leading hypothesis (race condition, timing, selector/view issue, app bug, environment, shared state, data flakiness).
- Test it in a way that can reject it; confirm or reject; record the confirmed cause and the evidence.
- Goal: a confirmed root cause, not a guess.
Phase 4: Fix & Validation
- Decide the fix (test-side vs product-side) and, critically, the alternatives you rejected and why.
- Apply it, then validate stability over repeated runs.
- Goal: a stable, verified fix with a documented decision.
Phase 5: Prevention
- Decide how to prevent recurrence: a shared helper, a lint rule, documentation, a regression guard.
- Record the preventive action(s).
- Goal: the failure does not come back silently.
File Purposes
Scale the file scope to the investment level (triaged at the start — see Flow). Higher value = fuller record; lower value = leaner:
| Investment | Files in project root | How much to write |
|---|---|---|
| P1 high-value / blocking | All three: plan + findings + progress | Full pipeline: goal, phases, decisions, errors, run log |
| P2 medium | plan + findings | Phases and the why; progress only if the session runs long |
| P3 low-value / cosmetic flake | findings only | Evidence + classification + suspected cause; move on |
Each investigation creates the files above in the project root:
| File | Purpose | When to Update |
|---|---|---|
qa_investigation_plan.md | Goal, phases, decisions, error log | After each phase completes |
qa_investigation_findings.md | Root cause, evidence, technical decisions | After ANY discovery |
qa_investigation_progress.md | Session log, run/result records | Throughout the session |
Critical Rules
- Create the plan first — non-negotiable; the plan is your persistent memory. For a P3 (low-value) case, the
findingsfile is the plan — create that first. - 2-Action Rule — after every 2 read/search ops, save key findings to
qa_investigation_findings.md. - Read before decide — re-read the plan before major decisions.
- Update after act — mark phase status, log errors, note files changed.
- Log ALL errors — with attempt number and resolution.
- Never repeat failures — if an action failed, the next must differ.
- Classify after reproducing, not before — a wrong early label poisons the investigation.
