test(codascon): close all open test gaps (§17 multi-hook, §18 strategy lifetime, §_CSS3 callable proof)

[scorpevans]

Addresses all open gaps in the codascon test suite identified in the test analysis.

Summary

• §17 — Multi-hook H tuple at runtime: adds two it blocks for RichFeedCommand extends Command<..., [LogCommand, GroomCommand]>. Both hooks are invoked inside execute; the result depends on both hook return values, so the assertion only passes when both hooks fire. A second it verifies hooks are re-invoked (not cached) on each run() call.

• §18 — Strategy statefulness (cached vs fresh): adds two it blocks contrasting CachingCommand (holds a CountingStrategy instance as a field — state accumulates across run() calls) vs FreshCommand (creates a new CountingStrategy on every resolveDog call — state resets). Confirms the framework is neutral about strategy lifetime.

• _CSS3 — CommandSubjectStrategies callable proof: adds a callable proof to constraints.test.ts. cmd.run(dog, { time: "x" }) compiling proves that a correctly implemented FeedCommand satisfies its own this & CommandSubjectStrategies<...> constraint. If any visit method were missing or had the wrong signature, tsc would reject the call site.

  Limitation (documented in typescript-gotchas.md): CommandSubjectStrategies is an internal (non-exported) type used as the this constraint on run(). It cannot be exported with @internal — stripInternal: true would strip the declaration while leaving the reference in run()'s signature, producing a dangling name in the published .d.ts. Shape assertions (_CSS1/_CSS2) were attempted and reverted for this reason. _CSS3 + the @ts-expect-error blocks in §14 together close the gap.

Test plan

• pnpm --filter codascon test — 46 tests pass (36 runtime + 10 compile-time)
• pnpm --filter codascon build — tsc succeeds; CommandSubjectStrategies stays as private declaration in .d.ts, no dangling reference

🤖 Generated with Claude Code

[scorpevans]

Reviewed. §17 adds two it blocks. First: food === 'morning' requires groomResult.groomed (GroomCommand invoked) AND logEntry.action (LogCommand invoked) — neither hook can be dropped without breaking the assertion. Second: running with different time values shows hooks are called fresh on each run(), not memoised. CommandHooks<[LogCommand, GroomCommand]> structural requirement (log + groom properties) is enforced at compile time by tsc. No existing tests touched. 44/44 pass.

[scorpevans]

Reviewed. Verified: §17 both-hooks invoked and result depends on both (groomResult.groomed selects branch, logEntry.action provides value); §18 cached vs fresh strategy instances behave as expected; _CSS1/_CSS2 correctly use the base class form to work around the circular-inference limitation (documented); _CSS3 callable proof compiles; all 46 tests pass and build clean.