Date: 2026-11-08 Scope:
main
in one go before Phase 47’s TFP-247 publishes an image.Status: Complete 2026-11-12: merged to main as 2861a4c (2026-11-11);
the first scheduled nightly on main green (run 43191298031); v0.3.0 tagged on
49431af and pushed, with the owner’s confirmation (Decision 13), and
released: v0.3.0
(2026-11-12).
Locked (Emil, 2026-11-08). In progress on branch
phase-48-mr-onion: TFP-275 done (46e3994), TFP-276 done (621dd8a);
TFP-277 done (9844233); Gate A independently verified and passed
(Emil, 2026-11-09). TFP-278 done (48604dc, baseline 8892b7e),
independently verified (2026-11-09). TFP-279 done (d4f10cf, c35217d). TFP-280 done (276f7b5); its
allowlist accepted at Gate B (Emil, 2026-11-11).
Gate B passed (Emil, 2026-11-11); TFP-281 (the repo rename and merge)
next.
Tracking issue #25; draft PR
#23.
mr-onion
spelling, the one-time data-folder move, the old names working until 1.0,
and no domain for now (Decisions 2, 6, 7 and 12).What carries the name today (main at 4459382, 2026-11-08):
package.json: "name": "testforge",
"private": true (nothing is published to npm), "version": "0.2.0",
"bin": { "testforge": "bin/testforge.mjs" }, and an npm script
testforge. bin/testforge.mjs starts src/cli/main.ts and passes
TESTFORGE_CLI_CWD. src/cli/config.ts has the usage text
(testforge run) and the default --out of ./testforge-out.v0.2.0 is tagged under the current name (D39 item 5).src/lib/seo.ts (SITE_NAME, SITE_TITLE, which
feeds layout.tsx’s page titles and Open Graph), the header in
src/app/page.tsx:330, src/app/_components/HelpPanel.tsx, the report
pages (src/lib/report/html.ts: titles, header and two messages;
comprehensive-html.ts), and the report footer (src/lib/report/text.ts:
“TestForge buildJUnit(report, name = "TestForge") in
src/cli/junit.ts: the <testsuites name>, and the fallback suite name
for a record with no suite or area.TESTFORGE_* names are read across src, scripts,
bin, tests, .github and .env.example, most often _INTERACTIVE,
_DATA_DIR, _RUNS_ROOT, _OLLAMA_MODEL, _GAP_LIMIT and
_CLAUDE_COMMAND. The run’s child processes get some as an internal
protocol: _REQUEST_CAP_URL/TOKEN/DOMAIN, _NOT_JUDGED_TITLES/REASON,
_STORAGE_STATE, _PROGRESS (the reporter’s stdout marker in
executor.ts).// Generated by TestForge ... (mock/index.ts,
comprehensive.ts);test.skip(process.env.TESTFORGE_INTERACTIVE !== '1', '... TESTFORGE_INTERACTIVE=1 ...')
(database-round-trip.ts, journey-tests.ts, gap-fill.ts,
comprehensive.ts, through INTERACTIVE_ENV_VAR);testforge-missing-…, the values TestForge check <time>, the sample
testforge@example.com, TestForge-Sample-1, TestForge sample,
testforge-check@example.com, the fixture files
testforge-drop-fixture.*, data-testforge-probe and
window.__testforgeXss.src/lib/site-suites/reuse.ts, with writtenHash catching hand edits).
A saved suite written today goes on checking
process.env.TESTFORGE_INTERACTIVE after the rename unless it is
regenerated.src/lib/storage/data-dir.ts: TESTFORGE_DATA_DIR,
else %LOCALAPPDATA%\TestForge, ~/Library/Application Support/TestForge,
or $XDG_DATA_HOME/testforge (~/.local/share/testforge). It holds runs,
site suites and candidates. --baseline latest and run.json read from
it (Phase 47 Decision 7 amended, TFP-246).ollama/Modelfile.testforge-qwen7b defines the tag
testforge-qwen7b, the TESTFORGE_OLLAMA_MODEL example value
(.env.example, docs/LOCAL_MODEL_SETUP.md). Each user builds it on
their own machine. The system prompt opens “You are TestForge”
(src/lib/generate/prompt.ts)..claude-plugin/marketplace.json (testforge-skills-marketplace,
plugin testforge-skills) and skills/.claude-plugin/plugin.json.emil3663/testforge, public. Workflows: codeql.yml
(push and PRs to main, weekly), pr-validation.yml (every PR),
self-test.yml (push to main, manual), regression-test.yml (manual),
nightly.yml (schedule, manual; TEMPLATE_BASELINE: 06489fd; secret
TESTFORGE_SMTP_PASSWORD). A scheduled workflow runs only on the default
branch. Links to github.com/emil3663/testforge appear in README.md,
TEST_PLAN.md, CHANGELOG.md, .github/ISSUE_TEMPLATE/config.yml and
four decision files.emil3663.github.io address, so there is nothing to redirect.main’s README from the
branch root, now at emil3663.github.io/mr-onion/ (its title read
“TestForge | mr-onion” after the rename, before the merge). Pages
doesn’t redirect after a rename, so emil3663.github.io/testforge/
no longer answers (404). Nothing tracked links to either address.
Acceptance Criteria 7 checks the title after the merge.testforgeapp@gmail.com, with +ci for the nightly email
(TFP-217) and +<site> for site runs (D25).benchmarks/, examples/), dated records
and tests.phase-16-local-first, feat/napi,
docs/claude-skills-catalog.mr-onion; settings use MRONION_.
(Emil, 2026-11-08)
emil3663/mr-onion; package mr-onion; command mr-onion
(mr-onion run); bin file bin/mr-onion.mjs; image
ghcr.io/emil3663/mr-onion (TFP-247); plugin mr-onion-skills,
marketplace mr-onion-skills-marketplace; --out default
./mr-onion-out.TESTFORGE_X becomes MRONION_X, the rest of the name
unchanged. Code identifiers use mrOnion / MrOnion where a name is
needed.%LOCALAPPDATA%\MrOnion,
~/Library/Application Support/MrOnion, $XDG_DATA_HOME/mr-onion
(~/.local/share/mr-onion).mronion everywhere, which gives one spelling but reads as
“mro-nion”. A hyphen is valid in every place mr-onion goes; it isn’t
valid in environment variable names, hence MRONION_.seo.ts), report pages, report footer and CLI messages.bin/ entry and the npm script.README.md, CONTRIBUTING.md, SECURITY.md, CHANGELOG.md (a new
entry; past entries keep the name they were written under),
THIRD_PARTY_NOTICES.md, docs/ living docs (the “Docs that must stay
true” table in WORKING_AGREEMENTS.md, plus docs/README.md,
LOCAL_MODEL_SETUP.md, AUTOMATED_ACCOUNT_SETUP.md,
GENERATOR_COMPARISON.md and APPLICATION_OVERVIEW.md), CLAUDE.md,
the decisions index, the roadmap and the register’s open items.<testsuites name> and its fallback suite name become
“Mr Onion”.TESTFORGE_* setting, including the child-process protocol
names and the TESTFORGE_PROGRESS marker.Mr Onion check <time>,
sample addresses, fixture names, the probe attribute and the window
global). A site owner reading their logs sees the tool’s real name.
The template check proves only the name moved (Decision 11).ollama/Modelfile.mr-onion-qwen7b and default tag
mr-onion-qwen7b. When that tag is missing and testforge-qwen7b is
present, the old tag is used, and the generation record says so
(ollama cp testforge-qwen7b mr-onion-qwen7b is the documented fix,
with no rebuild).tests/regression/
(testforge.*.spec.ts), so the leftover-name check needs no allowance
for them.TFP- ticket prefix, and every test-case and test ID prefix
(TFFU-, TFSE- and the rest, the test plan’s area prefixes).
Reports, baselines and history refer to them.benchmarks/ and examples/ (frozen evidence, never edited:
WORKING_AGREEMENTS.md, “Where names still belong”); past CHANGELOG.md
entries; git history and the v0.2.0 tag and release.run.json), site suites, candidates
and library entries, which keep the text they were written with
(Decision 8).testforgeapp@gmail.com and its +ci and
+<site> aliases: the nightly email’s sender display name becomes
“Mr Onion nightly”. A new mailbox is D44.source\repos\testforge, WSL’s ~/testforge-wsl): they aren’t
user-facing, and renaming them breaks the owner’s tooling paths for no
gain. Changing the remote URL is enough (Decision 10).testforge stays a second bin entry, and npm run testforge stays a
script. Both print one line on stderr: “testforge is now mr-onion;
the old name stops working at 1.0.” Then they run the same CLI, with
the same exit code.MRONION_X, then TESTFORGE_X. When the old
name supplied the value, the app prints one notice per process naming
both, on the server log or CLI stderr. When both are set and differ,
the new one wins and the notice says so.
Amended 2026-11-09 (Emil, owner decision): a blank MRONION_X
(empty or only spaces) doesn’t hide a TESTFORGE_X that has a value:
the old value is used, and the notice says the new name is blank. A
pipeline that templates in an empty new variable keeps its old one. A
blank MRONION_X on its own stays as it was: returned blank, which
each reader treats as unset.MRONION_DATA_DIR, TESTFORGE_DATA_DIR or
--data-dir) is never moved: it is used where it is.run.json, run folder or site suite written by 0.2.x loads, diffs
and serves as --baseline <id> and --baseline latest unchanged.
Nothing in their format is renamed.TESTFORGE_INTERACTIVE. Every run’s child environment sets each
protocol and gate setting under both names, with the same value. An
old saved suite therefore runs exactly as before, interactive tests
included.writtenHash’s hand-edit detection, and edits
promotion evidence).nightly.yml reads secrets.MRONION_SMTP_PASSWORD || secrets.TESTFORGE_SMTP_PASSWORD.
The owner adds the new secret in GitHub’s settings (an owner action,
outside the repo), and deletes the old one after the first green
nightly on main sends the email. Both steps are confirmation
required.Amended 2026-11-08 (Emil): the secret is not renamed in this
phase. TESTFORGE_SMTP_PASSWORD is only the label GitHub stores the
mailbox’s app password under; it isn’t user-facing, and the mailbox
itself stays (Decision 4). It keeps its name until a new mailbox is
decided (D44), so there is no fallback expression and no owner step in
GitHub’s settings. check:old-name’s allowlist carries it, with this
reason. The sender display name still becomes “Mr Onion nightly”.
emil3663/testforge to emil3663/mr-onion in GitHub’s settings
(confirmation required); Claude Code runs
git remote set-url origin https://github.com/emil3663/mr-onion.git
(repo-local config) and merges at once.emil3663/mr-onion. They resolve from
the moment of the rename, and old links and clones redirect.emil3663/testforge: this is
recorded in SECURITY.md’s maintainer notes and in the register.main’s
docs would say “TestForge” under a repo called mr-onion); renaming
after the merge (main would briefly link to a repo name that doesn’t
exist yet).check:template-unchanged gains --rename-map <file>, a committed
list of literal old → new replacements
(scripts/template-unchanged/rename-map.json). It applies the map to
the base ref’s output, then compares byte for byte with the head’s.
Exit 0 only when every fixed input is identical.TestForge → Mr Onion,
TESTFORGE_ → MRONION_, testforge- → mr-onion-,
__testforge → __mrOnion and so on). A token that isn’t a name is
a review finding.mronion.com serves a live site, and mronion.dev
answers (robots.txt present). Both look taken; a registry lookup
wasn’t possible from the planning session. A listed Taiwanese company,
Mr. Onion Corp. (catering, TPEx 2740), joins the Chinese restaurant
chain already in D39’s trademark notes.mr-onion.dev,
mronion.app, mr-onion.app, mr-onion.io, mronion.io, and
mronion.co.za for the home market..dev or .app, both
HTTPS-only TLDs, which suits a security-minded tool.ghcr.io/emil3663/mr-onion and never a testforge
image. Outside pipelines pin the image name, so it must be right the
first time (D39 item 5).mr-onion
(image ghcr.io/emil3663/mr-onion, command mr-onion run); they start
after Phase 48 merges.”v0.3.0 is tagged on
main after the merge, and the tag push is confirmation required, as
with v0.2.0.main meanwhile. [for Emil]
phase-48-mr-onion, from main at the lock. The name follows
phase-16-local-first.main is opened with the first ticket, so
PR Validation and CodeQL run on every push. It is marked ready only at
Gate B.main gets no code
commits while the branch is open: Phase 47 is paused, and Dependabot
PRs wait. Docs-only commits from planning sessions may land on
main.main is merged into the branch (a merge, never a
rebase or force-push: those are confirmation required).main is a merge commit (--no-ff), so each
ticket’s Conventional Commit and its evidence stay in main’s
history. Rejected: squash (loses the per-ticket reasoning the commit
bodies carry); rebase-merge (no single commit to point at or revert).main, unchanged. On the
branch, the nightly is dispatched by hand (workflow_dispatch, ref
phase-48-mr-onion) once after TFP-278’s baseline commit and once at
Gate B, both green.check:template-unchanged against the lock’s main: every fixed input
identical, with no map.process.env.TESTFORGE_* read remains outside the settings module
(a grep, quoted).MRONION_X, and under TESTFORGE_X alone
with exactly one notice per process. With both set and different, the
new value is used and the notice says so. Unit tests show each case,
seen failing first.mr-onion run and testforge run give the same exit code, JUnit and
report.json (apart from run id and timings) against local Toolshop
clean and with-bugs. testforge prints the deprecation line on stderr,
and mr-onion prints nothing extra.%LOCALAPPDATA%\TestForge present (runs, a site
suite, a candidate): the first start renames it to MrOnion, prints
the notice once, and the UI lists the same runs. --baseline latest
takes the same run as on 0.2.0. A saved site suite is reused, not
regenerated.
Amended 2026-11-09 (Gate A; the criterion was written from Phase
47’s context line, which the code doesn’t bear out): the default
data folder holds site suites, candidates and gap-writer settings,
not the UI’s runs. Runs are in the repo’s runs/ (RUNS_ROOT)
unless the runs folder is set inside the data folder, as --data-dir
does. So the criterion reads: the UI’s candidate list is unchanged,
and a run kept in the data folder (via --data-dir) serves as
--baseline latest after the move. The repo’s runs/ isn’t touched.~/.local/share/testforge → mr-onion), in the
WSL clone. macOS is covered by unit tests only, stated as a limit.--data-dir) is never
moved.run.json diffs against a new run and serves as --baseline
with the same known/new/fixed split as before the rename.check:template-unchanged <before> <after> --rename-map …: every fixed
input identical. Without the map, the raw diff is attached, and every
changed line is a name token.MRONION_INTERACTIVE=1 and skips it without (Decision 8).TEMPLATE_BASELINE is re-recorded in nightly.yml in its own commit.
A dispatched nightly on the branch is green.testforge-qwen7b present, local generation uses it, and the
generation record names the fallback.--help and error output, and junit.xml contain
no “testforge” in any case. The exceptions are the deprecation notice,
a pre-0.3.0 run’s footer (Decision 7) and the target’s own content. A
script renders each one and greps it; its output is quoted.<title>, og:site_name).<testsuites name> is “Mr Onion”, and a record with no suite
falls back to “Mr Onion”. The counts equal report.json’s, as in Phase
47 criterion 2.mr-onion-skills from the renamed
marketplace, and its skills load.npm run check:old-name greps tracked files case-insensitively and
exits 0 only when every hit is in its allowlist
(scripts/old-name/allowlist.json). Each entry is a path or pattern
with a reason, and it covers only: dated records and reports in
docs/decisions/, benchmarks/, examples/, past CHANGELOG.md
entries, the compatibility code and its tests, the rename map,
the nightly’s TESTFORGE_SMTP_PASSWORD secret name (Decision 9,
amended), and LICENSE’s absence of the name. The allowlist is reviewed line by line
at Gate B.CHANGELOG.md’s 0.3.0 entry states the new names, the old names’
end at 1.0, the data-folder move, the plugin reinstall and the rollback
note (Decision 14).main merged in: every criterion
above re-checked by an independent verifying session; unit,
self-test and regression suites in full; PR Validation and CodeQL
green; a dispatched nightly green.main: the merge commit’s tree is identical to
the branch’s last verified commit, plus nothing (git diff --stat
empty, quoted). The scheduled nightly is green and its email arrives
(Decision 9). The old URL github.com/emil3663/testforge redirects
(web and git ls-remote). The template check against the branch’s
baseline is identical. check:old-name passes.
Added 2026-11-11 (Emil): once the Pages build after the merge
finishes, the <title> of emil3663.github.io/mr-onion/ reads
Mr Onion (“Mr Onion | mr-onion”, the second half being the repo
name), with no old name, quoted.docs/test-plan/, with TEST_PLAN.md as the index, in the same
commit: the command and its old name in CI.md; the generated text and
the mapped template check in GEN.md; the data folder, saved runs and
baselines in the area that already holds run persistence and diffing.
Existing case IDs keep their prefixes (Decision 4).behavior-preservation-verification skill). It adds no
cases of its own.QUALITY_GATES.md §1): the
fallback reads with the fallback removed, the move with the move removed,
and the mapped check with a mutant.Each is added to deferred-work-register.md in the same change as this
lock:
TESTFORGE_* names are regenerated
or rewritten. Until then, runs set both names (Decision 8). Not before
1.0’s removal ticket, and not as part of it.testforgeapp@gmail.com and its
aliases, if one is wanted (Decision 4). The nightly’s secret is renamed
with it (Decision 9, amended 2026-11-08).refactor, renameprocess.env.TESTFORGE_* across src, scripts and bin, so a
fallback would have to be repeated at every read.src/lib/settings.ts, which exports setting(name) (the
short name, e.g. "DATA_DIR") and the list of known names. Every
product read goes through it. scripts/ and tests/ reads too, where
they read a product setting. No new names, and no behaviour change.46e3994): check:template-unchanged cf501ad
172 of 172 identical with no map; unit 791 of 791, self-test 5 of 5,
regression 73 of 73, none skipped. Accepted (Emil, 2026-11-08): the
direct reads left outside src/lib/settings.ts:
src/lib/run/request-cap-fixture.ts:27-31 (copied into each run folder
and run in the Playwright child, so it can’t import the module; its env
comes from the parent), scripts/nightly-email/notify.mjs:41 (the
nightly’s secret, which keeps its name: Decision 9, amended), and the
save/restore around writes in src/app/api/generate/route-test-helpers.ts:47,72.feature, rename, cliTESTFORGE_* and type
testforge; both must keep working while the new names take over.setting() reads MRONION_X then TESTFORGE_X, with the
notices of Decision 5. The child environment sets both names
(src/lib/run/child-env.ts, request-cap.ts, auth.ts,
interactive.ts, the reporter marker). bin/mr-onion.mjs plus the
testforge alias, the npm scripts, usage text and --out default.
.env.example lists the new names. The generated suite’s text is
untouched.feature, rename,
storage…\TestForge; the new default must find them without the user doing
anything.src/lib/storage/data-dir.ts: the new defaults
(Decision 2), the move and its cases (Decision 6), the notice, and the
footer’s version rule (Decision 7, src/lib/report/text.ts).9844233): verified at Gate A, passed (Emil,
2026-11-09). Every check used
copies under a temporary LOCALAPPDATA or XDG_DATA_HOME; the real
%LOCALAPPDATA%\TestForge was unchanged (41 files, no MrOnion)
before and after each. Evidence: 9844233’s body, RUN-50 to RUN-53.
For the verification and the owner, not decided by the build:
runs/
(RUNS_ROOT), which the move doesn’t touch. The build’s check put
a run in the folder (--data-dir), showed it serving as
--baseline latest after the move, and showed the UI’s candidate
list unchanged;package.json is still 0.2.0, so every footer on the branch says
“TestForge 0.2.0” until the 0.3.0 bump, which no ticket names;main then needs TESTFORGE_DATA_DIR
pointed at the moved one (Decision 14’s rollback note).generator, renameINTERACTIVE_ENV_VAR
and skip reasons, probe paths, test data, fixtures, probe attribute,
window global, sample addresses), the system prompt, the Modelfile and
default tag with fallback, and --rename-map in
scripts/check-template-unchanged.mjs with the map file. Then
TEMPLATE_BASELINE in its own commit.48604dc; TEMPLATE_BASELINE 8892b7e),
independently verified the same day. The mapped template check 172 of 172 identical; raw, 307
changed lines, each the map applied; a one-character mutant red; a
0.2.0 saved suite’s interactive test Not run without --interactive
and run with it; the old local tag used, and named, when only it is
built. The nightly dispatched on the branch at 8892b7e: run
42556213111, every job green, its email sent. Evidence: 48604dc’s
body, GEN-109 to GEN-112.Verified 2026-11-09 (planning session, independently). Every checked part of Acceptance Criteria 4 met:
| Check | How | Result |
|---|---|---|
| Commits exist and are pushed | git log, git show --stat, git ls-remote origin |
48604dc, 8892b7e, 5bd1dde on phase-48-mr-onion; the remote branch is at 5bd1dde; main is still cf501ad. 8892b7e changes only TEMPLATE_BASELINE (06489fd to 48604dc) |
| The map holds names only | Read rename-map.json |
Six tokens, each a spelling of the name (TestForge-Sample-1, TestForge, TESTFORGE_, __testforge, testforge-, testforge@) |
| Mapped check | check:template-unchanged 3c742e8 --rename-map …, in a fresh clone |
172 inputs, 172 identical, 0 different, exit 0; replacements 2, 222, 52, 8, 51 and 2 |
| Raw check | The same, with no map | Exit 1, every input different |
| Only the name moved | An independent script over the raw outputs: each changed line compared with every spelling of either name blanked | 307 changed lines, 0 with any other difference; the old name, in any case, 0 times in the head’s output (172 inputs in the base’s) |
| The mapped check can go red | A mutant: one character of the generated header comment changed (key. to key!) |
2 identical, 170 different, exit 1, the line named; reverted in the scratch clone |
| Saved suites keep running | Read interactive.ts, child-env.ts’s runEnv, playwright-cli.ts |
Every run child gets the switch under both names; the UI’s switch shows for a suite that checks either name |
| Old local model tag | Read resolveLocalModel and the generation gate |
Only when the new tag is asked for and missing; the record’s “model tag” verdict names ollama cp |
| Nightly on the branch | Run 42556213111’s page | Nightly #36, workflow_dispatch on phase-48-mr-onion at 8892b7e: success; unit, audit, dead-code, template-unchanged, issue and notify all ran |
Not checked here: the live saved-suite run against local Toolshop and
the full unit, self-test and regression suites (no browser or Windows
shell in this session; they rest on 48604dc’s body, and Gate B
re-runs the suites). Accepted as reported: the old-name branch of the
timing filter has no test, since a probe route is always written fresh.
ui, report, renameseo.ts, page.tsx, HelpPanel.tsx, report/html.ts,
comprehensive-html.ts, report/text.ts, CLI messages,
junit.ts’s default, package.json’s name, marketplace.json,
plugin.json, workflow step names, tests/regression/ file names. Plus the render-and-grep script of
criterion 5.
Amended 2026-11-09 (Emil, Gate A ruling 2): also package.json’s
version to 0.3.0, so the footer of a run on the branch names Mr Onion
and criterion 5’s report check can pass. main doesn’t see 0.3.0 before
the merge (Decision 14). The changelog’s 0.3.0 entry stays in TFP-280,
and the v0.3.0 tag in TFP-281, after the merge (Decision 13).d4f10cf; the plugin enabled under its new name in
c35217d; PR #23’s checks green on c35217d). Built 2026-11-09. check:visible-name on
the built app and a Standard and a Comprehensive run: no old name
found (RPT-22); JUnit named Mr Onion with counts equal to report.json’s
(CI-23); the template check 172 of 172 identical against 48604dc, no
map; package.json at 0.3.0 (Gate A ruling 2). The plugin installs as
mr-onion-skills from the renamed marketplace with its 12 skills,
under a temporary CLAUDE_CONFIG_DIR (the owner’s condition, 2026-11-11;
the real config unchanged), and the repo’s .claude/settings.json
enables the new name. Evidence: d4f10cf’s and c35217d’s bodies.docs, rename, ciCHANGELOG.md’s 0.3.0 entry under
Unreleased; repo links to emil3663/mr-onion; check:old-name with its
allowlist, added to PR Validation; CLAUDE.md’s one-line rule (“the
product is Mr Onion; check:old-name must pass”). The rule for current
text against history: present-tense text is renamed; a sentence that
records what happened on a date, a quoted output or a frozen path keeps
the old name.
Amended 2026-11-11 (Emil): the 0.3.0 entry’s plugin reinstall note
says what a tester with the old plugin does. Once their checkout has
the renamed marketplace.json, testforge-skills@testforge-skills-marketplace
no longer resolves (“not found in marketplace”). They uninstall it,
add or update the marketplace (now mr-onion-skills-marketplace), and
install mr-onion-skills@mr-onion-skills-marketplace; the repo’s
.claude/settings.json enables the new name (TFP-279). The skills are
the same twelve.
Corrected 2026-11-11 (Gate B verification; the build tested it):
“add or update the marketplace” is wrong. An update keeps the old
registered name, and the new plugin then isn’t found. The tester removes
testforge-skills-marketplace and adds the checkout again, as the
changelog’s four commands do.276f7b5; the Gate B doc fixes in e50cc6e; the
allowlist accepted at Gate B, Emil, 2026-11-11, criterion 6 read as
ruled there; PR #23’s checks green on e50cc6e). Built 2026-11-11. check:old-name: 16,838
hits in 1,175 files (tracked paths included), all allowed by 52 of 52
entries; seen failing on a planted name, a stale entry and an entry
with no reason; in PR Validation (TST-08). For Gate B to rule on:
CLAUDE_CONFIG_DIR, an update keeps the old
registered name, and the new plugin then can’t be found; the
marketplace has to be removed and added again. The changelog gives
those four commands.main — Gate Brelease, renamemain must go from TestForge to Mr Onion in one
step, proven on the merged result.main into the branch; independent verification of
criteria 1 to 6; the owner’s repo rename and the remote URL (Decision 10);
the --no-ff merge; the post-merge checks; v0.3.0.
Amended 2026-11-11 (Gate B verification; Decision 9 as amended
2026-11-08): there is no secret switch-over. The secret keeps its
name, so the post-merge check is only that the nightly’s email
arrives.
Amended 2026-11-11 (Emil): also the Context section’s Pages line
corrected, and the Pages site’s title checked after the merge
(Acceptance Criteria 7).v0.3.0 tag and release, once confirmed.2861a4c, 2026-11-11; the owner’s go-ahead,
2026-11-11). v0.3.0 tagged on 49431af and released 2026-11-12,
with the owner’s confirmation (Decision 13).
origin set to
https://github.com/emil3663/mr-onion.git after the owner’s rename.
PR Validation on 0c25416 failed before lint: GitHub’s
ubuntu-latest had moved to ubuntu-26.04, and “Playwright does
not support chromium on ubuntu26.04-x64” (run 43048248301, both
attempts). By the owner’s choice (2026-11-11), 83c7d87 pins every
job to ubuntu-24.04, and TFP-284 (in the workflow-and-CI backlog)
tracks the Playwright upgrade. On 99903a3, PR Validation (old-name
step: “16842 hits in 1175 files; 16842 allowed by 52 of 52 entries”)
and CodeQL were green. The dispatched nightly 43049335228 passed on
attempt 3: unit 821 tests, 820 passed, 0 failed, 1 skipped (the
Windows-only shell test). Attempts 1 and 2 failed before any test
ran, because the runner couldn’t resolve azure.archive.ubuntu.com;
they opened issue #26, which attempt 3 closed. main hadn’t moved
since the lock (cf501ad, the merge base).git merge --no-ff phase-48-mr-onion, 2861a4c, with
parents cf501ad and 99903a3; PR #23 shows as merged.| Post-merge check (Acceptance Criteria 7) | Result |
|---|---|
| The merge’s tree is the branch’s last verified commit | git diff --stat phase-48-mr-onion 2861a4c: empty |
| Old URL redirects | Web: 301, Location: https://github.com/emil3663/mr-onion; git ls-remote https://github.com/emil3663/testforge.git returns 2861a4c for main |
Template check against 48604dc |
“172 inputs, 172 identical, 0 different”, exit 0 |
check:old-name |
“16842 hits in 1175 files; 16842 allowed by 52 of 52 entries”, exit 0 |
| Pages title | <title>Mr Onion \| mr-onion</title>, built from 2861a4c (2026-11-11 17:25 UTC); the old emil3663.github.io/testforge/ returns 404 |
| Smoke Test and CodeQL on the push | 5 passed (2.0 min); CodeQL success |
| Scheduled nightly and its email (Decision 9) | Run 43191298031 (schedule, main, 2861a4c, 2026-11-12 09:25 to 09:44 UTC): all six jobs green; unit 821 tests, 820 passed, 0 failed, 1 skipped (the Windows-only shell test); notify sent to testforgeapp+ci@gmail.com; arrival confirmed by the owner |
| Template check in the scheduled nightly | “48604dc -> 2861a4c: 172 inputs, 172 identical, 0 different” |
2861a4c); Phase 48 Complete in this record,
the roadmap and the index; the changelog’s 0.3.0 entry dated
2026-11-12; issue #25 closed.phase-48-mr-onion and its draft PR.v0.3.0.mr-onion.Gate A verification (planning session, 2026-11-09): every checked criterion met. The owner’s ruling: passed (Emil, 2026-11-09). The planning session verified it independently. It designed the record but built none of TFP-275 to TFP-277.
| Check | How | Result |
|---|---|---|
| Commits exist and are pushed | git log, git show --stat, git ls-remote origin |
506c5b5, 9844233, 69142d0, 81e239a on phase-48-mr-onion; the remote branch is at 81e239a; main is at cf501ad |
| Scope | git show --stat |
The move, the store’s old-path mapping, the footer rule and their tests; plus the self-test server’s temporary data folder (a real isolation defect found while building, fixed in the ticket per the standing policy: kept) |
| Template unchanged (criteria 1-3) | check:template-unchanged cf501ad, in a fresh clone of the branch |
172 inputs, 172 identical, 0 different, exit 0 |
| Data-dir, settings and version tests | Re-run, Linux, Node 22, temporary XDG_DATA_HOME/LOCALAPPDATA |
34 passed, 0 failed, 1 skipped (needs Chromium) |
| The move on a real filesystem | A probe calling dataDir() twice per process, in temporary folders |
Old only: renamed, file intact, one notice. Both: new used, old left, one notice. Rename refused (read-only parent, EACCES): old used, notice gives the reason; the next start moved it. MRONION_DATA_DIR or TESTFORGE_DATA_DIR set: that folder used, the default old folder not moved (the old name’s deprecation notice printed) |
| The owner’s real folder | Listing of %LOCALAPPDATA% |
TestForge present, no MrOnion |
| Code read against Decision 6 | data-dir.ts, store.ts, text.ts |
Matches, including the two review fixes (a process left on the old folder follows a later move; a pre-move record’s suite path maps to the new folder) |
Not checked here, with where they are checked:
EPERM with a held file,
--baseline latest’s 0 new / 46 known / 0 fixed on both versions, 470
reused) and TFP-276’s live two-command run: this session has no Windows
shell or browser. They rest on 9844233’s and 621dd8a’s bodies, with
their outputs quoted;69142d0 and 81e239a: GitHub couldn’t be read
from this session. Confirm they are green before TFP-278 starts.Rulings:
%LOCALAPPDATA%\TestForge before running the branch himself, then lets
it move as it will for every user after the merge. Claude Code’s checks
keep using copies under a temporary folder.mainmain merged in,
independently verified; the allowlist reviewed by the owner.Gate B verification, part 1: the branch (planning session, 2026-11-11). Every criterion checked here is met, except the three doc findings below, which are fixed in the working tree. The planning session built none of TFP-279 or TFP-280. The owner’s ruling: pending.
| Check | How | Result |
|---|---|---|
| Commits exist and are pushed | git log, git show --stat, git ls-remote origin |
d4f10cf, c35217d, 7d59b4e, 276f7b5, d9a7853 on phase-48-mr-onion; the remote branch is at d9a7853 |
main merged in |
git merge-base main phase-48-mr-onion |
cf501ad, which is main’s head: main hasn’t moved since the lock, so the branch already contains it |
| Package | package.json |
"name": "mr-onion", "version": "0.3.0" |
| Leftover names (criterion 6) | npm run check:old-name, in a fresh clone |
16,838 hits in 1,175 files, all allowed by 52 of 52 entries, exit 0 |
| The allowlist | All 52 entries read | Each has a reason, and each falls in one of five groups: frozen evidence and dated records (1-9); compatibility code and its tests (10-36); the owner’s decisions to keep a name (37-40); history lines inside living files, matched line by line (41-52). Nothing user-facing found behind an entry. Whole-file entries are used only for compatibility code and dated documents |
| Template, no map (criteria 4-5) | check:template-unchanged 48604dc |
172 inputs, 172 identical, exit 0 |
| Template, mapped (criterion 4) | check:template-unchanged 3c742e8 --rename-map … |
172 inputs, 172 identical, exit 0 |
| Types | tsc --noEmit -p . |
exit 0 |
| Changelog’s 0.3.0 entry (criterion 6) | Read | New names, old names until 1.0, the data-folder move, the plugin reinstall (four commands), the repo, and going back to 0.2.x: all present |
Findings, fixed in the working tree by the planning session (doc edits):
CHANGELOG.md said “Your runs, site suites and candidates move with it”
and that going back lets 0.2.x find “your runs”. The app’s runs are in
the repo’s runs/ folder and don’t move (Gate A ruling 1). Both lines
now name site suites, candidates and gap-writer settings.Not checked here, and where they are checked: the full unit, self-test
and regression suites, and npm run check:visible-name (it needs the
built app and real Standard and Comprehensive runs). This session’s VM
has no Chromium, and each unit file takes about 30 s to start here. A
fresh Claude Code session on the owner’s machine that built none of
Phase 48 runs them as part 2. Also not checked here: PR #23’s checks on
d9a7853 (reported green).
For the owner’s ruling:
docs/consolidation/, the dated handover and briefs, the four owner
decisions, history lines in living files). Recommended: accept it, and
read criterion 6 as “frozen evidence, dated documents, compatibility
code, the owner’s decisions to keep a name, and history lines in living
files, each with its reason”.
Ruled 2026-11-11 (Emil): accepted. Criterion 6 reads as above.Gate B verification, part 2: the suites (a fresh Claude Code session on
the owner’s machine that built none of Phase 48, 2026-11-11). It ran on
e50cc6e (only docs changed since d9a7853), with a temporary
LOCALAPPDATA and no data-folder setting. The real
%LOCALAPPDATA%\TestForge was identical before and after (41 files), and
nothing was committed.
| Command | First run | Rerun, nothing changed |
|---|---|---|
npm run test:unit |
821 tests: 820 passed, 1 failed, 0 skipped (18 min 18 s), exit 1 | The failing file alone (routes.test.ts): 3 of 3 passed, exit 0 |
npm run test:self |
“Timed out waiting 120000ms from config.webServer”, 0 tests ran, exit 1 | 5 of 5 passed, exit 0 |
npm run test:regression |
73 passed, none skipped, flaky or failed, exit 0 | — |
npm run check:visible-name |
“no old name found”, exit 0. Title “Mr Onion | …”, og:site_name “Mr Onion”, JUnit <testsuites name="Mr Onion"> with counts equal to the reports’, the footer “Mr Onion 0.3.0 (e50cc6e)” |
— |
The two first-run failures, read by the planning session:
data/labelled-failures.md). The rename touched
only name strings in that test and the code it runs. Ticketed as
TFP-282.Gate B ruling (Emil, 2026-11-11): passed, green on the reruns. Each
first-run failure is logged as one data point with its cause. If either
fails the same way again, it is treated as a real signal, not explained
again. Next: the owner renames the repo (Decision 10), then the
--no-ff merge and Acceptance Criteria 7’s post-merge checks.