Version: 1.0
Date: 2026-09-23
Status: Complete. TFP-078/079 955601e, TFP-080 abb94f0, TFP-082 6f90a0e (all 2026-09-23). TFP-081’s three pieces landed separately: the help panel in 8be5dd7, the README update in e231a8b, and .clinerules is in .gitignore. Reconciled with git 2026-09-29.
Sprint 3 (Phase 24, TFP-065-070) landed and was verified ticket-by-ticket
against the actual repo, not just the report — all four remaining commits
(a95027a, 2bd1b32, f54e9e1, 3f3e7ac) check out. Re-running the
project’s own npm run lint as part of that verification surfaced 15
pre-existing errors that none of those four commits introduced (confirmed:
none of the flagged files appear in any of the four diffs), plus a handful
of other loose ends accumulated across the last few sessions. Emil’s call:
resolve all of it before picking the next roadmap item, rather than
carrying it forward.
Nothing here is new functionality — this phase is entirely cleanup, config correction, and finishing verification that was legitimately deferred (not skipped) in Sprint 3.
no-require-imports correctly: exclude skills/ from the app’s eslint configtech-debt, configeslint.config.mjs’s ignores list excludes
build artifacts (.next, out, build, etc.) but not skills/, so
skills/testing/plan-to-specs/*.js gets linted under
eslint-config-next’s ESM-only ruleset and fails
@typescript-eslint/no-require-imports on 4 lines across 3 files. This
is a scoping bug, not a style violation: those files are standalone,
#!/usr/bin/env node-shebanged CommonJS scripts, shipped as part of the
testforge-skills plugin for someone else’s repo to run directly with
plain Node — they were never meant to be linted as if they were part of
this Next.js app’s own ESM source tree.skills/** to eslint.config.mjs’s ignores array.src/ is silently being linted
under the same mismatched assumption (a quick check, not a full
audit — this backlog isn’t asking for a general eslint-scope review
beyond what’s already been found).eslint.config.mjs.npm run lint no longer reports the 4
no-require-imports errors; skills’ own tests (if any exist for
plan-to-specs, e.g. npm test inside that directory, or however that
skill runs its own suite) still pass — this is a lint-scope fix, not a
code change to the skill itself.actionable-elements.ts along its existing function boundariestech-debt, refactorsrc/lib/generate/mock/actionable-elements.ts is
517 lines, 17 over this repo’s own max-lines: 500 cap (scoped to
src/**/*.ts(x), and enforced for real — TFP-068 extracted
pollRun.ts specifically to respect it). Unlike dom-snapshot.ts
(see TFP-080), this file has no page.evaluate() serialization
constraint — it’s imported normally by demoqa-suites.ts,
generic-suites.ts, and link-audits/index.ts — and already has 5
clearly separable top-level exported functions
(buildActionableElementTests, buildToggleGroupTests,
buildAccessibilityElementTests, buildUsabilityElementTests,
buildActionPairTests), so a real split is low-risk here, not just
theoretically possible.activeAriaStateField/elementWords with whichever exported
function actually uses each).actionable-elements.ts (or update the three
importing files directly) so nothing calling into this module today
has to change its import path unless that’s the cleaner outcome —
Code’s call, state which was chosen and why.behavior-preservation-verification — prove output is unchanged via
an actual before/after fixture diff on generated test output, not by
reading the diff and asserting it looks equivalent.npm run test:unit and the existing regression suite both still pass;
the fixture-diff proof is included in the commit message or as a
committed artifact, not just asserted.dom-snapshot.ts’s max-lines exceptiontech-debt, configsrc/lib/introspect/dom-snapshot.ts is 817
lines, but it is not a normal candidate for extraction the way
TFP-079’s file is. The file’s own existing comment (lines 10-14) states
why: captureDomSnapshot is passed by reference to Playwright’s
page.evaluate(), which serializes the function’s own source and runs
it inside the browser’s isolated JS context with no access to this
module’s scope — every helper it uses must stay nested inside its body,
because there is no module system or import mechanism available at
runtime in that context. Splitting it into separate files, however
clean-looking in the diff, would silently break real page introspection
the moment it runs against a live page, since the extracted helpers
would no longer be reachable inside the serialized function. Treat the
existing in-file comment as authoritative — verify the constraint is
still real (check dom-snapshot.ts’s own call site for
page.evaluate(captureDomSnapshot, ...) or equivalent) rather than
taking the comment itself as sufficient proof, then fix the tooling
gap it exposes rather than the file.page.evaluate() call site and that it does pass this
function by reference (not by re-implementing its body as a string
or similar) — this is the one fact this ticket needs verified
firsthand, not assumed from the comment.max-lines override for this one file (an
eslint.config.mjs block keyed to its exact path, or a targeted
/* eslint-disable max-lines */ at the top of the file with a
comment pointing back to the existing rationale) — whichever this
repo’s existing eslint-disable conventions (if any) already favor.captureDomSnapshot’s internals as part of this ticket.README.md’s Known Limitations section (or wherever the doc-audit’s
earlier work landed similar notes) stating this file is intentionally
long and why, so a future contributor doesn’t rediscover this from
scratch.npm run lint no longer flags
dom-snapshot.ts; every other file under src/** is still held to the
500-line cap (this is a named, single-file exception, not a general
loosening of the rule).choresrc/app/_components/HelpPanel.tsx
and the page.tsx wiring, independently verified (tsc/eslint clean,
all five content sections present, dismiss behavior confirmed).README.md doc update (Skills library section, the honest-
coverage and run-diffing bullet updates) from the earlier
repo-claim-audit pass..clinerules was agreed (Cowork session, 2026-09-23) to be added to
.gitignore rather than committed as a tracked file — that
.gitignore entry was never actually added; .clinerules is still
untracked..clinerules to .gitignore in its own small commit.docs/HANDOVER_TO_CLAUDE_CODE_2026-09-22.md untracked and
uncommitted — that exclusion was deliberate (it contains a stale
claim, see that session’s notes) and still stands.git status is clean except for
docs/HANDOVER_TO_CLAUDE_CODE_2026-09-22.md, which stays untracked on
purpose.reliability, testing:3000 hit an unrelated Turbopack/Tailwind internal
error mid-session and was correctly left alone rather than restarted
without authorization. It’s still down. Emil wants nothing left
unresolved before moving on, and this is the one piece of Sprint 3
that’s genuinely still open, not just under-verified.next dev on :3000 (not touched or diagnosed yet — this
ticket is where that actually happens).next build && next start
(or an equivalent that doesn’t collide with the shared .next
directory the way the commit message flagged) confirming the actual
served response headers omit unsafe-eval in production.next dev runs cleanly on :3000 with no
startup errors; TFP-068’s retry/backoff is observed live, not just
proven at the unit level; TFP-069’s production CSP header is confirmed
from an actual served response.npm run lint and npx tsc --noEmit are both fully clean, repo-wide.git status is clean except for the one deliberately-untracked
handover doc.