feat(packs): re-land the pack lint + the EE/sensing domain packs (#1328)

[aarontrowbridge]

Part of amicode #1328. Plan-20260920 step 10 re-landed fresh off current main — supersedes #1329 (closed stale-base per the #1465 doctrine; branch 1328-domain-packs preserved).

What lands (byte-identical replay of ab1260fb onto main 244e6a9c)

Zero conflicts — main's Fleet Studio reshaping (09-23/24) touched none of this slice's paths (packages/schema, packages/extension/src/scores, packages/extension/packs).

• The schema extension (additive; the flagship stays valid): the domain keys — curricula, payload_schemas, verification (contract reference with claim-class bindings), instruments (each entry carrying its coverage scope bound — modeled / modeled-at-first-task / unmodeled-until-hardware), benchmarks, first_task (the shakedown | substrate level ladder). scores minItems relaxed deliberately: a seeded domain pack legitimately carries no composed scores yet — the quantum-era single-pack contract, not the general one.
• The pack lint: schema validation plus reference resolution (dangling refs = named findings) plus the seeded-key check. The flagship reads formalized, not seeded — exactly the grandfathered design — and the lint surfaces a known real finding baked into the test: the flagship's skill paths are snapshot-layout paths that dangle in-repo; the step-14 formalization reconciles them, and this test keeps that from landing silently.
• The two seeded packs, each with: the verification contract (tier assignments, claim-class bindings — hardware claims un-tiered until measured), the payloads doc (every payload carrying the tier enum), the first-task declaration (shakedown-marked, NOT yet run — exercised is a run state, never a manifest state), a curriculum skeleton authored to be filled from measured practice, and corrector integrity manifests (load-time shas, the WS1 — The pack interface: one manifest per domain, score as a field of a pack #369 pattern).

Verified (from package dirs only; mirrors the #1329 record verbatim)

• packages/schema: bun run test — 284/284 passed (20 files)
• packages/extension: bunx vitest run test/pack_lint.test.ts test/packaging.test.ts test/gate_packs.test.ts — pack_lint 4/4; packaging 3 passed (+4 skipped); gate_packs 58 — 65 passed | 4 skipped total; EE + sensing lint seeded: true, the flagship formalized with the known finding class
• packages/extension: bunx vitest run test/scores — 323/323 passed (20 files); includes the bundled-packs loader check loadPacks([packs]).errors === [] (all three packs load clean)
• bun run typecheck — clean in both packages/schema and packages/extension

Honest scope notes

• A manifest is cheap; a domain is not (the r3 advisory, honored): the exercised bar — first task run end-to-end, tier-labeled verdict — is the next milestone per pack, tracked in each FIRST-TASK.md.
• Two authoring bugs caught by running, not inspection (in the original campaign): the TOML table-header ordering and the loader's corrector-integrity requirement.

Resolves #1328.

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Advanced

Run ID: 92b252fc-4033-4b46-9f6d-c3264fc6b54f

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}