qa-bug-repro

Action-taking agent that ingests a Playwright trace.zip (and/or a HAR file plus console logs and screenshots) and emits a structured bug report matching the `bug-report-template` schema - verbatim error messages, repro steps reconstructed from the recorded actions, environment block populated from the trace metadata, expected-vs-actual filled from the recorded assertions and observed responses. Distinct from `bug-report-template` (which fills the same template from chat / voice / informal text) and from `bug-repro-builder` (which converts the report into a failing test). Use when a manual tester or CI run captured a Playwright trace and the team needs a triageable issue without round-tripping for missing fields.

Action-taking agent that turns a bug report into a minimal failing test (or, when the bug needs more context than a unit test allows, a minimal repro repository). Reads the bug report, identifies the smallest unit of code that exercises the failure, generates the failing test in the project's test framework, and runs it once to confirm it actually fails. Use immediately after triage to lock in reproduction and create the regression-prevention asset.

Read-only agent that parses a crash dump or stack trace, identifies the crashing frame and the source line, attributes the failure to a suspected commit range using `git blame` on the relevant file, and emits a root-cause hypothesis. Handles JS/TS V8 traces, Python tracebacks, Java/JVM stack traces, Go panics, native (gdb/addr2line) traces, and minified production stacks (with sourcemap support). Use when the only input is an error log or crash report and you need a starting point for repro.

Read-only agent that groups a backlog of bug reports into root-cause clusters using stack-trace overlap, summary text similarity, and shared error fingerprints. Returns a cluster table with each cluster's representative bug, member count, and shared signal. Use when triaging a long bug backlog (50+ open issues) to find duplicates and identify the few root causes producing the bulk of the noise.

Workflow orchestrator that runs a full weekly defect review in sequence: clustering (defect-clusterer) -> trend narration (defect-trend-narrator) -> escape analysis (escape-defect-analyzer), then assembles one consolidated report for the QA manager or lead. Distinct from each sibling agent run in isolation: this agent owns the hand-off contracts between stages, enforces the required inputs at each step, and refuses to proceed when upstream output is missing. Use when a QA manager or lead wants a single, end-to-end weekly defect review rather than running each specialist agent ad hoc.

Builds a well-formed bug (defect) report from raw observation notes - fills in summary, environment, steps to reproduce, expected vs actual, and severity rationale. Validates that each field has the load-bearing content reviewers and engineers need to triage. Use when a stakeholder reports a problem informally (chat, email, voice) and the team needs a triageable issue without round-tripping for missing fields.