Version: 1.0 Date: 2026-09-16 Status: Shipped — content built out under Phases 18-21 (skills/README.md, skills/ROADMAP.md are the live source of truth); this doc’s own header was never updated when the work landed. Corrected 2026-09-23 (Cowork verification pass) — see the Sept 22 handover note that first flagged this.
docs/REIMAGINING.md proposed rebuilding TestForge as a set of testing
skills for coding agents instead of a standalone app. That proposal was
written against main-origin, the pre-Phase-16 codebase (a 4,739-line
page.tsx, a 2,226-line run/route.ts, a 2,415-line generate/route.ts,
and a FastAPI backend). Phase 16 already deleted the FastAPI backend
entirely and decomposed all three of those files — page.tsx is 353
lines with no code editor, run/route.ts is 120 lines, and the actual
prompt content already lives on its own in src/lib/generate/prompt.ts
(29 lines). The literal “delete ~12,000 lines” reading of the proposal no
longer applies to a codebase that’s already been decomposed and is about
to ship as test-assist.
What still holds from the proposal, reconciled against Emil’s own
answers to Martin’s four questions: TestForge is paused, not abandoned,
so this is additive work alongside the app, not a replacement for it
(Q1). The prompt + problem statement are worth packaging now, since the
prompt is already isolated and more skills are expected to surface as
work continues (Q2). Claude Code is the only target for now; portability
to Copilot/Cursor is an explicitly deferred follow-up (Q3). Testing is
multi-disciplinary — this backlog treats skills/ as a growing set, and
no single skill is positioned as the product (Q4).
skills/ directory at the repo root.REIMAGINING.md as candidates, not commitments.introspect/crawl currently do, before anyone commits to
rebuilding that as a skill.src/app or src/lib — the shipping app is untouched.honest-coverage, dont-stub-what-you-prove,
find-the-hidden-platform, or plan-to-specs — these are recorded as
candidates only in this phase.enhancement, skills, claude-codeSYSTEM_PROMPT — is currently only
reachable by running the TestForge app. Packaging it as a Claude Code
skill lets that judgment be used directly inside a coding agent’s own
workflow, without installing or running TestForge, and without
touching the app Phase 16 just finished decomposing.skills/write-playwright-test/ directory, additive only.SKILL.md built from the existing content of
src/lib/generate/prompt.ts (SYSTEM_PROMPT +
BEST_PRACTICE_GUIDANCE), rewritten for a skill’s instruction format
rather than copied verbatim as an LLM system-prompt string.skills/write-playwright-test/SKILL.mdskills/README.md stating this is the first of several planned
skills, not a rename of the product.git diff shows zero changes outside skills/ and the new README.generate endpoint currently applies.app.py, or the deleted proxy layer appears
anywhere in the new skill.documentation, planningskills/ROADMAP.md listing the four other candidate skills from
REIMAGINING.md (honest-coverage, dont-stub-what-you-prove,
find-the-hidden-platform, plan-to-specs) as candidates.skills/ROADMAP.mdskills/ is a growing set and no
entry is marked primary or flagship.investigation, spikesrc/lib/introspect and
src/app/api/crawl/route.ts are roughly 1,200 lines of real DOM
exploration logic — the one part of the app REIMAGINING.md flagged as a
genuine loss if deleted outright. Before any find-the-hidden-platform
-style skill gets built, confirm whether Playwright MCP already
provides equivalent capability, so that logic isn’t duplicated and then
maintained twice.src/lib/introspect and crawl/route.ts currently handle.skills/ROADMAP.md or linked from it) on the
overlap and gap between Playwright MCP and TestForge’s own
introspect/crawl logic.find-the-hidden-platform
skill needs new exploration logic or can lean on Playwright MCP.