Files
hyperframes/.github
James Russo 0c05025c04 ci(skills): run skills/**/*.test.mjs in CI (#1724)
skills/**/*.test.mjs files (e.g. skills/media-use/scripts/resolve.test.mjs
and skills/media-use/scripts/lib/manifest.test.mjs) are bare `node --test`
files with only `node:` built-in imports. They aren't part of any workspace
package, and the existing `Test` job's path filter (the `code` filter in
the `changes` job) excludes `skills/**`, so even on PRs that touch only
skills/ those tests never run.

This matters for regression guards. The shell-injection probe test added in
HF#1723 feeds probe() a filename containing `clip"; touch INJECTED; echo
".mp4` and asserts no marker file is created. The test passes locally but
under the current job graph it would never run in CI on a follow-up skills/
change that re-introduces the bug.

Closing the gap with a dedicated `Test: skills` job rather than relaxing
the `code` filter. The existing `Test` job's steps run `bun run test:scripts`
(hardcoded file list) and `bun run --filter '*' test` (workspace packages
only), neither of which would actually execute skills tests even if the
filter let `skills/**` through. The dedicated job needs no `bun install`,
just node 22, since the tests only import from `node:` and relative paths.

The discovery step shells out to `find` and fails loudly when zero test
files match, so a future rename or layout change can't silently turn this
into a no-op pass.

Spotted by Via in HF#1723 review thread, confirmed by James as a separate
follow-up rather than a blocker for HF#1723.

--
Jerrai (https://claude.com/claude-code)
2026-06-25 12:50:43 -07:00
..