skill-drift lint: hardware-loop fence scope resolves Intonato, pin expects Strumento (fleet-gated red, CI-blind)

[aarontrowbridge]

Important

Problem — The skill-drift lint's "real fleet regressions" battery has one test that fails on any fleet machine with live Harmoniqs Julia checkouts, at main tip (verified on the hub at the #1177 merge). The test pins that the hardware-loop skill's fence scope is Strumento (comment-stripped), but the live skill's fence resolves to Intonato. The test is mount-gated on the presence of real package checkouts, so CI never runs it — the red is fleet-machine-only and invisible to the merge gate.
Approach — Adjudicate which side drifted: either the skill's fence was re-scoped to Intonato without updating the pin (then the fence is wrong — the hardware-loop claims are about the Strumento/QICK board path — and the skill fence needs repair), or the re-scope was deliberate and the test pin is stale (then update the pin with a note). The #1002 false-positive lineage and the hardware-loop skill's edit history decide.

Acceptance Criteria

• On a fleet machine with canonical package checkouts (e.g. the hub), the full skill-drift-lint battery passes — including the "live hardware-loop regression: all ten fence claims VERIFY" test — with zero skips beyond the documented mount-gated ones.
• The adjudication (fence wrong vs. pin stale) is recorded in the closing comment, so the fence-scope convention has a durable precedent.
• CI remains green (the test stays mount-gated; no attempt to make CI depend on live checkouts).

Reproduction

Run the skill-drift-lint test suite on a machine with live checkouts under the canonical packages root. Fails with: expected [ 'Intonato' ] to deeply equal [ 'Strumento' ] at the fence-scope assertion inside the "live hardware-loop regression" test. Passing state: the same suite on CI (mount absent → skip) and both other fleet-regression tests on the hub.

Prior Art

• The skill-drift lint battery "real fleet regressions (amicode#1002 lint false positives)" — the ext-scan gap test and the hardware-loop test share the mount-gated runner-existence pattern.
• The hardware-loop skill's fence block and its Strumento scoping history (the QICK board path it documents).
• amicode#1002 — the false-positive regression class this battery pins.

Source

Found during the 2026-09-15 fleet-sync nightly build (macbook, main @ #1177): the suite's fleet-gated tests are the only main reds with CI blind to them.

Notes

Two sibling reds seen on the same run are environmental, not main reds (verified against the hub): the ext-scan-gap VERIFY goes DRIFTED against stale per-machine package checkouts, and the chat-bridge tilde test hits a Node-23/darwin rmSync cleanup quirk (its product assertions pass). Neither blocks this issue; the ext-scan one argues for a documented "fleet machines must run repo-sync before gating" rule if it recurs.