Version: 1.0 Date: 2026-09-16 Status: Shipped — TFP-040/041/042 all landed against test-assist (examples/testforge-app/test-assist/, examples/skills/write-playwright-test/test-assist/, examples/comparison-test-assist.md all present and referenced throughout later phases). This doc’s header was never updated when the work landed. Corrected 2026-09-23 (Cowork verification pass).
Emil asked for two related things: once skills exist, actually use them
rather than leave them as documentation — and specifically, run
TestForge’s actual app against test-assist to see what it produces,
plus separately run the skills themselves directly against
test-assist’s codebase, as a standing pattern that repeats for each
new skill as it lands (not a one-off).
test-assist turns out to have two builds, and only one is reachable by
anything TestForge does:
npm run serve (port 4321) or live at
https://emil3663.github.io/test-assist/. Already has 55 of its own
hand-written Playwright tests (a smoke suite in CI, a regression suite
on demand) — a real baseline to compare against, not just eyeball.write-playwright-test. This is out of reach for this whole phase,
not a gap to work around — stated plainly so it isn’t assumed later.This phase only targets the browser build.
Running TestForge’s own app requires a live dev server (and, for
TFP-040, test-assist’s browser build served at the same time) staying
up across multiple steps — that needs Claude Code’s persistent terminal,
not Cowork’s stateless shell, which can’t keep a background process
alive between tool calls.
test-assist’s browser build, served locally.write-playwright-test directly against the same target, with no
TestForge app involved — skill judgment in isolation.test-assist’s own existing 55-test suite.examples/ as a standing location this repeats into for every
future skill, not just this one run — matching the examples/
folder REIMAGINING.md originally proposed (“a real suite built with
them, warts recorded”).test-assist’s desktop (PySide6) build — unreachable by this
tooling entirely.test-assist itself.dogfooding, integrationtest-assist’s browser build locally: npm run serve in
test-assist (port 4321).examples/testforge-app/test-assist/ — the generated suite, the
run report, and a short note on what worked and what didn’t.
Warts recorded, not cleaned up before showing.dogfooding, skillswrite-playwright-test (via Claude Code, no TestForge app
involved) directly against test-assist’s browser build — its
source (app.js, index.html) and the served instance.examples/README.md stating this pattern repeats for
each future skill (code-structure, refactor-audit once
promoted; the SRE/ISTQB candidates once built) against the same
target, so there’s a consistent baseline across skills over time.examples/skills/write-playwright-test/test-assist/examples/README.mddogfooding, analysistest-assist’s
existing hand-written 55-test suite. Without a direct comparison,
having three teaches nothing.examples/ — no new tooling.examples/.