Stats
Actions
Tags
From skills
Runs exploratory QA in two phases — phase 1 generates and runs many breadth probing scripts derived from the codebase (route/HTTP sweeps, smoke flows, malformed-input fuzzing) headless; phase 2 deep-dives the live app in a real browser via Playwright (MCP/CLI), following curiosity. Produces a ranked findings report. Use this whenever the user wants exploratory testing, a bug hunt, a pre-release sweep, or to "poke at the app and see what breaks".
How this skill is triggered — by the user, by Claude, or both
Slash command
/skills:exploring-qualityThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Scripted tests only find bugs you already imagined. Exploratory QA finds the rest — by
Scripted tests only find bugs you already imagined. Exploratory QA finds the rest — by
sweeping the surface broadly, then chasing the anomalies deep. Work in two phases: breadth
first (cheap, headless, many probes), then depth (a real browser, curiosity-led) on whatever
phase 1 flagged. The output is a findings report, not permanent tests — keepers get handed to
authoring-tests, bugs to tracking-issues.
Derive the app's surface from the code, then probe it widely and fast.
scripts/surface-inventory.sh # routes, SSE events, forms, slash-commands from the code
scripts/launch-demo.sh & # start the offline demo server
scripts/breadth-sweep.sh # HTTP status sweep + malformed payloads + rapid SSE
Look for: non-2xx where you didn't expect it, 500s on malformed input, unescaped output, slow paths, broken redirects, missing headers. Collect every anomaly with its repro. See references/phase1-breadth.md.
Take phase 1's leads (and your own curiosity) into the live browser via the bundled
playwright-cli / Playwright MCP. Drive real flows — streaming, type-ahead queueing,
permissions, plan mode, elicitation — using accessibility-tree snapshots, not pixels.
Run axe-core as you go, and capture a screenshot for anything visual so the finding has
evidence. See references/phase2-deepdive.md.
Write a ranked report (docs/qa/exploratory-<date>.md) using
references/findings-report.md: each finding has severity, a
repro, evidence (screenshot/log), and a suggested home — a tracking-issues issue and, if it
should never regress, an authoring-tests test.
Breadth without depth gives you a pile of status codes with no story; depth without breadth means you deep-dive the first thing you see and miss the rest. Sweep wide to know where to look, then look hard only there — that's how you cover a lot of surface without burning hours.
The demo server (./my-orchestra -demo) and the vendored playwright-cli skill already exist.
Note the sandbox can't bind servers — run phases locally or in CI, not inside a restricted tool call.
npx claudepluginhub dotts-h/claude-skills --plugin skillsProvides CDSS development patterns for drug interaction checking, dose validation, clinical scoring (NEWS2, qSOFA), and alert classification integrated into EMR workflows.