Split The code is public; the app answers only to the crew.
The tree guard is red on any machine that has run the app: it fails on files .gitignore already excuses #129
commented
Reproduced again at 4161986, one command, from a clone whose check was green a minute earlier:
$ mkdir -p data && touch data/app.db && bun test src/tree.test.ts
(fail) the repository holds no cookie jar and no scratch directory
0 pass, 1 fail
Removing data/ makes it pass again. openDatabase defaults to ./data and the README names data/app.db, so this is not an exotic state — it is what the app does the first time anyone runs it outside the forge. The guard should fail on what is committed (and on a cookie jar anywhere), not on what the app wrote five minutes ago; git ls-files is the list it wants to walk.
Sign in to join this conversation.
new above
offline
src/tree.test.tslanded with #98 and does the right thing about cookie jars. But it flags any file whose name matches a glob in.gitignore, and an ignored file is proof the ignore works, not a violation. The result: the two workflows the README documents both turn the check red on a developer's own machine.Reproduced at
ea3cc6f, in a clean clone with nothing modified and nothing staged:openDatabasedefaultsFORGE_APP_DATAto./dataand the README's Run it locally section says the database lands atdata/app.db, so running the app once makes the check fail until you delete its database.bun run builddoes the same throughstatic/app.*.js, which.gitignorealso lists. CI is green only because a fresh clone has neither.Done looks like
Files I expect to change:
src/tree.test.ts, and nothing else.Risks:
git ls-filesis not available to the test (the header says so), so "committed" has to be approximated — skipping ignored paths is the honest approximation and is what the acceptance criteria ask for; do not reach for agitsubprocess, which is a shell built from paths and is refused by the bar; and keep the walk cheap, since it runs on every check.Brief that fits: scout — the test and its intent.