style: remove em/en dashes repo-wide, add CLAUDE.md rule against them (#890)

Em dashes (—) and en dashes (–) had spread across comments, docs, tests,
and a few UI strings, reading as AI-generated boilerplate rather than
house style. Replaced each with punctuation matching its context: colon
for explanatory clauses, comma for asides, plain hyphen for numeric/legal
ranges (e.g. "21-23§"), "to"/"till" for date ranges, parentheses for
paired-dash asides. messages/en.json and messages/sv.json were fixed by
hand together to keep sv/en in sync.

Left untouched where the dash is the functional subject rather than
decorative punctuation: date-range-parser.ts's separator regex,
charset-repair.ts's CP1252 byte-mapping table (and its test), the SIE
encoding mojibake docs, generic-csv.ts's minus-sign normalizer, the
agent system-prompt files that already instruct against em dashes, and
a golden iXBRL test fixture compared byte-for-byte.

Also fixes two bugs surfaced along the way: an off-by-one in
ApiKeysPanel's scope-label split (a leftover from an earlier partial
pass), and a charset-repair test that had lost the literal en-dash it
exists to verify.

Regenerated the agent atom seed migration (skills:generate) since 27
SKILL.md files changed. Added a CLAUDE.md rule against em/en dashes,
with an explicit carve-out for the functional-dash cases above.

Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
This commit is contained in:
Jakob Wennberg
2026-07-04 15:58:06 +02:00
committed by GitHub
co-authored by Claude Sonnet 5
parent a5154ee884
commit ec27228a8e
1588 changed files with 28388 additions and 10890 deletions
+9 -9
View File
@@ -4,23 +4,23 @@
*
* The audit found two repository-wide problems that are being remediated in
* dedicated campaigns (A1 = route auth/MFA, D1 = money rounding). Those touch
* hundreds of sites and won't land in one PR — so this guard makes sure the
* hundreds of sites and won't land in one PR: so this guard makes sure the
* count can only go DOWN, never up, while the migrations are in flight.
*
* Checks:
* 1. raw-route-auth — an `app/api/**\/route.ts` that calls
* 1. raw-route-auth : an `app/api/**\/route.ts` that calls
* `supabase.auth.getUser()` directly instead of going through
* `requireAuth()` / `withRouteContext()` (the only guards that enforce
* MFA AAL2 on hosted). Tracked as a file-set so a NEW offending route
* fails CI even if an old one was fixed in the same PR.
* 2. naive-ore-round — `Math.round(x * 100) / 100`, which is subtly wrong on
* 2. naive-ore-round: `Math.round(x * 100) / 100`, which is subtly wrong on
* exact-half values (see lib/money.ts `roundOre`). Tracked as a count.
* The canonical rounding modules are excluded.
* 3. direct-jel-insert — a file that inserts into `journal_entry_lines`
* 3. direct-jel-insert: a file that inserts into `journal_entry_lines`
* outside the sanctioned writers. During the dimensions dual-write window
* every line writer must derive cost_center/project via
* lineDimensionColumns() from the dimensions JSONB map
* (lib/bookkeeping/dimension-resolver.ts) — a new direct insert site can
* (lib/bookkeeping/dimension-resolver.ts): a new direct insert site can
* silently diverge the mirror columns. Tracked as a file-set.
*
* Usage:
@@ -37,7 +37,7 @@ const ROOT = path.resolve(path.dirname(fileURLToPath(import.meta.url)), '..', '.
const BASELINE_PATH = path.join(ROOT, 'scripts', 'checks', 'antipatterns-baseline.json')
const IGNORE_DIRS = new Set(['node_modules', '.next', '.git', 'dist', 'build', 'coverage'])
// The sanctioned home of the öre-round implementation — must not count against itself.
// The sanctioned home of the öre-round implementation: must not count against itself.
const ROUND_EXEMPT = new Set(['lib/money.ts', 'lib/bokslut/rounding.ts'])
const RAW_AUTH_RE = /\.auth\.getUser\(/
@@ -91,7 +91,7 @@ const JEL_INSERT_SANCTIONED = new Set([
'app/api/sandbox/seed/route.ts',
])
// Matches an insert CHAINED on the lines table (`.from('journal_entry_lines').insert(`,
// with optional whitespace/newlines in the chain) — select-only readers don't count.
// with optional whitespace/newlines in the chain): select-only readers don't count.
const JEL_INSERT_CHAIN_RE = /\.from\(\s*['"]journal_entry_lines['"]\s*\)\s*\.\s*(insert|upsert)\(/
/** Files that insert into journal_entry_lines outside the sanctioned writers. */
@@ -175,7 +175,7 @@ if (newAuthFiles.length) {
}
// 1b. direct-jel-insert: allowlist lives in this file (JEL_INSERT_SANCTIONED),
// no baseline — any unsanctioned insert site is a hard failure.
// no baseline: any unsanctioned insert site is a hard failure.
if (current.directJelInsert.length) {
failed = true
console.error(
@@ -210,7 +210,7 @@ if (fixedAuthFiles.length || current.naiveOreRound < baseline.naiveOreRound.coun
}
if (failed) {
console.error('\nAntipattern guard failed — see above.')
console.error('\nAntipattern guard failed: see above.')
process.exit(1)
}
console.log(
+8 -8
View File
@@ -5,13 +5,13 @@
* `npm run lint` was never wired into CI, so ~60 pre-existing errors
* accumulated across the repo. Fixing them all in one PR is churn; gating raw
* `eslint` would break every PR until then. So: ratchet. Error counts are
* tracked per rule in a committed baseline and can only go DOWN, never up —
* tracked per rule in a committed baseline and can only go DOWN, never up:
* a PR introducing a NEW error of any rule fails CI, while legacy errors are
* burned down independently.
*
* Warnings stay advisory (only `--quiet` errors are counted).
*
* Known tradeoff: counts are per-rule repo-wide, not per-location — a PR that
* Known tradeoff: counts are per-rule repo-wide, not per-location: a PR that
* fixes one legacy error of a rule can absorb one NEW error of the same rule
* without tripping the gate. Acceptable for a burn-down ratchet; tighten to
* per-file fingerprints if that ever bites.
@@ -36,7 +36,7 @@ function runEslint() {
['eslint', '.', '--quiet', '-f', 'json'],
{ cwd: ROOT, encoding: 'utf8', maxBuffer: 64 * 1024 * 1024 },
)
// ESLint exits 1 when errors exist — that's expected; only treat a missing/
// ESLint exits 1 when errors exist: that's expected; only treat a missing/
// unparsable report as fatal.
if (!result.stdout) {
console.error('no-new-lint-errors: eslint produced no JSON output')
@@ -77,7 +77,7 @@ if (process.argv.includes('--update')) {
BASELINE_PATH,
JSON.stringify({ totalErrors: total, perRule: sorted }, null, 2) + '\n',
)
console.log(`no-new-lint-errors: baseline updated — ${total} error(s) across ${Object.keys(perRule).length} rule(s).`)
console.log(`no-new-lint-errors: baseline updated: ${total} error(s) across ${Object.keys(perRule).length} rule(s).`)
process.exit(0)
}
@@ -97,7 +97,7 @@ for (const [rule, count] of Object.entries(perRule)) {
}
if (regressions.length > 0) {
console.error('no-new-lint-errors: FAILED — new ESLint errors beyond the baseline:\n')
console.error('no-new-lint-errors: FAILED: new ESLint errors beyond the baseline:\n')
for (const { rule, count, allowed } of regressions) {
console.error(` ${rule}: ${count} (baseline ${allowed})`)
for (const loc of (locations[rule] ?? []).slice(0, 10)) {
@@ -105,7 +105,7 @@ if (regressions.length > 0) {
}
}
console.error(`
Fix the new error(s) — run \`npx eslint . --quiet\` locally to see them.
Fix the new error(s): run \`npx eslint . --quiet\` locally to see them.
(If you fixed MORE legacy errors than you added and the rule still trips,
re-baseline with: node scripts/checks/no-new-lint-errors.mjs --update)
`)
@@ -114,8 +114,8 @@ re-baseline with: node scripts/checks/no-new-lint-errors.mjs --update)
const improved = total < (baseline.totalErrors ?? 0)
console.log(
`no-new-lint-errors: OK — ${total} error(s), baseline ${baseline.totalErrors}.` +
`no-new-lint-errors: OK: ${total} error(s), baseline ${baseline.totalErrors}.` +
(improved
? ' Count went DOWN — ratchet it: node scripts/checks/no-new-lint-errors.mjs --update'
? ' Count went DOWN: ratchet it: node scripts/checks/no-new-lint-errors.mjs --update'
: ''),
)