ci: make lint pass, so tests actually run

[dickhardt]

CI has failed on every branch since February. Because npm run lint runs before lerna run test, the test suite has not executed on a pull request in months — PR #65 merged in an UNSTABLE state for this reason. The publish workflow tests separately, so releases stayed gated; pull requests did not.

Two causes

Prettier had no ignore for generated output. .prettierignore listed node_modules, dist, archive*; the eslint config additionally ignored dist-test. The two lists disagreed, so prettier checked dist-test and .svelte-kit — which only exist after a build — and failed on 106 files nobody wrote. Aligned them and added .svelte-kit and .claude.

Eslint reported 14 errors. Seven from .claude/worktrees, a local directory it should never have read. The other seven were this idiom:

const { d, p, q, dp, dq, qi, ...publicJwk } = privateJwk

Naming the discarded fields is how the omission works — they are not unused. Set ignoreRestSiblings instead of scattering eslint-disable comments, which also retired the ones already in the tree.

One thing worth revisiting separately

Removing those disables reformatted the files they were in, which prettier then flagged. That is an ordering problem in the script:

prettier --check . && eslint --fix

The formatter check runs before the fixer, so anything eslint rewrites is never checked, and the failure surfaces on someone else’s next run. I left the script alone and committed the formatted result rather than change the lint contract in a fix PR — but --fix in CI mutates a checkout nothing will commit, so eslint without --fix is probably what CI wants.

Verified

prettier --check clean, eslint clean, npm run lint exits 0, and httpsig (145), better-auth and email-verification all pass.

🤖 Generated with Claude Code

https://claude.ai/code/session_01LXExbWHeem2SNb7tZ2LGnc