qa-flow plugin

Independently reviews a manual QA run — opens the screenshots, logs and scripts itself, attacks fragile PASS verdicts, finds missing coverage, and classifies each finding with the severity gate. Read-only; dispatched by the manual-test skill with a fresh context, never as a fork of the executor.

Executes an already-resolved manual QA plan against a running web app with a throwaway Playwright script, captures evidence, and returns a per-step PASS/FAIL report. Dispatched by the manual-test skill; never writes to the issue tracker and never asks the user questions.

Turn a task/ticket into a structured, professional QA test plan — a checklist of atomic, executable scenarios — and (optionally) post it back to your issue tracker. Use whenever the user asks to "create a test plan", "list test scenarios", "criar cenários de teste", "QA checklist for TICKET-XXXX", or otherwise wants a ticket turned into a verifiable QA plan. Works for any story type (UI, API, mobile, infra). Always runs a self-review pass before writing anything back.

File a bug (or feature/chore) ticket in your issue tracker via its REST API with a terse, budgeted description, attached screenshot evidence, owner/priority/placement filled in, and a verification GET before reporting done. Use whenever the user asks to "file a bug", "open a ticket", "criar um card", "abrir um bug", "report this to the tracker", or when manual-test hands over findings to file. Tracker-agnostic (Jira, Linear, GitHub Issues, Shortcut, ...).

Execute exploratory/manual QA against a web app, driven by either a ticket's test plan or a single ad-hoc scenario the user describes. Use whenever the user asks to "run a manual test", "fazer um teste manual", "execute this scenario", "validate this ticket manually", "manual test for TICKET-XXXX", or otherwise wants the agent to step through a behavior in the real app and report pass/fail with evidence. Resolves context with the user, delegates execution to an executor subagent, has the evidence checked by an independent reviewer, and only then (with approval) writes back to the tracker.

Turn ONE reviewed manual QA run (a throwaway Playwright script + its report and evidence) into a committed regression spec in the project's e2e suite — only if the scenario is strong enough to earn a permanent place, rewritten in the suite's own conventions, proven stable and proven able to fail, and independently reviewed before anything is committed. Use whenever the user asks to "promote this test", "promover para regressão", "turn this manual test into an automated one", "add this scenario to the suite", "automatizar esse cenário", or picks a candidate from scan-regression-candidates. Never commits or opens a merge request without the user's explicit OK.

Sweep a QA scratch root (default `_local/`, where manual-test keeps its throwaway scripts, evidence and logs) and list the few scenarios worth turning into committed regression tests — scored for value and cost, checked against the existing suite, with probes, one-offs and already-covered flows filtered out as noise. Read-only; produces a ranked backlog report and hands chosen candidates to promote-to-regression. Use whenever the user asks to "varrer a pasta local", "find regression candidates", "o que vale automatizar?", "which manual tests should become automated", "levantar testes pendentes de automação", or wants a backlog of automation from past manual runs.

Bash|PowerShell|Write|Edit|MultiEdit|NotebookEdit