git:20260918.95ebfff to git:20260921.5abb769

+ description: Automatically detects the target application's platform (Web/Desktop/Mobile/Hybrid/Embedded) and stack, dynamically selects the most effective E2E testing framework for it with a stated justification, and generates a Page Object Model plus a runnable E2E test script that asserts the application's own observable business behaviour — derived independently from the application's real features/flows and any flow description you give it. No fixed platform/language/framework list, and no dependency on /specclaw:bf-baseline or any captured Golden Master fixture — works standalone on whatever the target codebase actually is. Read-only with respect to application source — writes generated test/page-object files plus a persisted .specclaw/e2e/e2e-report.md, then (unless told not to) runs the generated suite once via specclaw-bf-e2e-run — starting any backend/frontend/dependency service it needs that isn't already running, in order, and leaving already-running ones alone — to record real Pass/Fail/Skipped counts, Pass Percentage and Execution Time in that report, plus on-failure screenshots/video collected into .specclaw/e2e/artifacts/ and listed there. Also renders a standalone .specclaw/e2e/e2e-report.html (Bootstrap via CDN) mechanically from the finished markdown report — a plain-language verdict banner, a pass-rate donut chart, stat cards, and a Test Scenarios list grouped by module/feature (never by raw file path) — one card per scenario, with the individual checks it makes nested underneath (so it's plain why an aggregate "32 passed" sits above fewer named scenarios) and the file path itself demoted to a small de-emphasized metadata line — plus inline screenshot/video evidence up front for any non-technical reader, with the developer-oriented detail (stack, run commands) tucked behind one collapsible "Technical Details" section — for stakeholders who'd rather open a styled page than read markdown. A one-off recorded run, not the authoritative verdict, which stays /specclaw:verify's job. Use whenever you need runnable E2E coverage for a target application's real user-facing flows.

+ This runs the agent's `install_cmd` from `.specclaw/e2e/run-config.json` (Task 6 of the agent), then brings up every declared background service — backend APIs, frontend dev/build servers, and dependencies like a database or queue — that its own `check_cmd` says isn't already running, one at a time in the declared order, waiting for each to report ready before starting the next. A service already running when this step starts is left exactly as it is; only services this step itself started are torn down again once `test_cmd` finishes. It then runs `test_cmd`, copies whatever on-failure screenshots/video the run produced (`run-config.json`'s `artifacts_dir`, wired up per the agent's Failure Capture task) into `.specclaw/e2e/artifacts/`, and mechanically patches the report's Execution Results and Artifacts sections — the real Pass Count, Fail Count, Skipped Count, Pass Percentage, Execution Time, and a listing of every captured file — never a number or a file the agent guessed. As its last step it renders `.specclaw/e2e/e2e-report.html` from the now-finished markdown — a mechanical transform, never a second place content is authored, so the two files can never disagree. This step has real side effects (it may install dependencies and start one or more services) and never fails the overall `/specclaw:bf-e2e` run: a missing precondition, a failed install, a service that never becomes ready, or an unparseable test-runner output is recorded in the report (and the HTML rendered from it) as "not executed" or "NOT MEASURED" with its reason, not treated as an error. Relay its one-line summary (or its "not executed" reason) to the user after the report-persistence line from Step 6, and mention the HTML report's path.