ci: run the TypeScript suite, and make its type checker pass (v1.43.1)

[lapc506]

Merge this before the rest of the open queue (#52, #41, #44, #59, #58, #46). Those six carry ~4,100 added lines that no CI would otherwise look at.

The gap

Three workflows, none running vitest or tsc:

Workflow	Trigger	Covers
test-hooks.yml	hooks/**, scripts/build-rules.mjs	346 shell tests
shellcheck.yml	**.sh	shell lint
publish.yml	push of a v* tag	npm run build

publish.yml effectively never fires — the note at the top of CHANGELOG.md records that this project shipped without v* tags. So src/, the published library, had no gate at all: 60 tests that ran when somebody remembered, and a tsc that had been red on three errors.

This is shape 1 from this toolkit's own ENF audit — "it does not look." And the partial coverage made it worse than an outright gap: the hook suite is gated, so every PR went green, and green reads as verified. A check covering part of a surface while reporting like it covers all of it is shape 4 applied to scope rather than to an assertion — correct, and binding nothing where it matters.

No paths: filter, deliberately

The one place test.yml departs from its siblings. A path filter is how this gap was built: scoping test-hooks.yml to the directory it tests is right for that workflow, and is exactly what left src/ uncovered.

Enumerating the paths that can break a type check means maintaining a second copy of the dependency graph, and the second copy is what drifts. The suite costs ~1.4 s. There is nothing to buy by scoping it.

Neither step is continue-on-error — an advisory test run is a correct verdict that stops nothing.

The type fix

printHumanResult demanded Record<string, unknown> while its body only calls Object.entries and never indexes by an arbitrary key. No *Result interface in src/lib/ declares an index signature, so none satisfied it.

-function printHumanResult(result: Record<string, unknown>): void {
+function printHumanResult(result: object): void {
...
-      printHumanResult(result as unknown as Record<string, unknown>);
+      printHumanResult(result);

Two lines. Fixes all three failing call sites and makes the cast at the fourth — the doctor path — unnecessary. Object.entries accepts both types identically, so no behaviour changes.

The cast is the artifact worth naming. Somebody hit this error at one of four call sites, silenced it with as unknown as, and the other three stayed red because nothing ever ran tsc. The fix is smaller than the workaround because the workaround never asked what the signature was demanding.

Verification

Run locally — exactly what the workflow will run:

tsc --noEmit          0 errors  (was 3)
npm test              60 passed
hooks/test-hooks.sh   346 passed

Not fixed here

The CHANGELOG link footer is missing entries for 1.39.0, 1.40.0, 1.41.0 and 1.43.0, and [Unreleased] still compares against v1.38.0. Four releases of drift, so it belongs in its own change rather than riding along here.

Created by Claude Code on behalf of @lapc506

🤖 Generated with Claude Code

[dojo-code-reviewer]

✅ Approved

Approved — 0 blockers, 1 P3. Confidence: 4.80/5.00.

Walkthrough

⚠️ Governance Warning: This PR targets main directly instead of following GitFlow expectations (feature → develop → main). This is a non-blocking informative notice, and does not affect the review verdict of the code itself.

Review Walkthrough

This PR establishes a TypeScript CI workflow (test.yml) running on every pull request and push to main to typecheck and run the Vitest test suite. It also fixes a compilation error in src/cli.ts by widening the parameter of printHumanResult from Record<string, unknown> to object, making the tsc compiler pass successfully.

Areas Reviewed

• CI Workflows: Reviewed .github/workflows/test.yml configuration, actions, and runner settings.
• CLI Code: Reviewed src/cli.ts for typings of printHumanResult and usage of Object.entries.
• Project Metadata: Reviewed .claude-plugin/marketplace.json, .claude-plugin/plugin.json, package.json, and CHANGELOG.md version bump consistency.

Safety Rationale

Widening the input type of printHumanResult to object has zero runtime impact and resolves static analysis errors due to missing index signatures on return types from src/lib/ functions. The test CI setup runs identical commands to those executed locally, ensuring continuous stability without impacting deployment or release flows.

Approved — 0 blockers, 1 P3.

🔵 P3 — Minor

• .github/workflows/test.yml:53 — 🔵 P3 (minor) — Adding a "typecheck": "tsc --noEmit" script to package.json and calling npm run typecheck here aligns CI commands with local developer tooling. Currently, developers must manually run npx tsc --noEmit which can drift from CI patterns over time if not declared as a standard package script.

[pass 1]

---

Total findings: 1 business context (1 total)

[dojo-code-reviewer]

🔵 P3 (minor) — Adding a "typecheck": "tsc --noEmit" script to package.json and calling npm run typecheck here aligns CI commands with local developer tooling. Currently, developers must manually run npx tsc --noEmit which can drift from CI patterns over time if not declared as a standard package script.

[pass 1]