test(generated): ts-rs export sync ratchet (#1132 PR-2)

[joelteply]

Summary

Ratchet test that locks shared/generated/<module>/index.ts barrels in sync with the .ts files emitted by ts-rs export tests. When they drift, types are invisible to TS consumers even though the .ts file exists on disk — exactly the regression that bit PR #1129 (commit db271d3 manually added 12 export lines after the fact). This makes that regression structurally impossible.

Scope (lane-separated with claude-tab-2)

PR-2 of continuum#1132. Lane separation acked on airc:

• claude-tab-2 took PR-1 (test(#1132 PR-1): canary smoke for AIRC + queue lifecycle #1135): AIRC + queue-lifecycle shell smoke slice (their territory — extends Paper: Qwen3.5-27B results #574/581)
• This is PR-2: Rust-side ts-rs export check slice (my territory — fresh from feat(persona): typed Engram + admission membrane types (#1121 PR-1) #1129/feat(persona): IsMemorable Recipe + admission gate (#1121 PR-2) #1134 ts-rs work)

Card

continuum#1136.

Test design

The ratchet:

• Walks every module dir under src/shared/generated/
• For each: lists .ts files on disk + parses index.ts for export type { X } from './Y' lines
• Asserts every on-disk file is referenced AND every reference has a file
• Failure message names the exact drift + suggests the npx tsx generator/generate-rust-bindings.ts recovery command

Critical correctness detail: extracts the FROM path (Y), not the exported type name (X). #[ts(rename = "...")] produces export type { ToolCall } from './AgentToolCall' where the file is AgentToolCall.ts but the type is renamed to ToolCall. First draft of the ratchet got this wrong and falsely flagged every rename usage on canary; corrected before commit.

Validation

$ cargo test -p continuum-core --features metal,accelerate --test generated_barrel_sync
test result: ok. 8 passed; 0 failed; 0 ignored

8/8 tests covering:

• 5 parser unit tests (canonical / rename / double-quote / multi-line / malformed-tolerance)
• 1 drift-detection unit test asserting both regression modes (missing-from-barrel + dangling-export) surface correctly
• 1 integration scan over the real shared/generated/ tree
• 1 known-modules sanity check guarding against accidental dir deletion hiding drift behind an empty module

npm run build:ts clean. Hooks ran without --no-verify.

Same shape as Lane F PR-1 deletion ratchet (#128) — walk source, assert pattern, fail loud with actionable message.

Test plan

• CI label-pr / validate / WIP green
• Future PR that adds a new #[derive(TS)] type without regenerating the barrel fails this test in CI (would have caught feat(persona): typed Engram + admission membrane types (#1121 PR-1) #1129 pre-merge)
• Future PR that deletes a .ts file but forgets to update the barrel fails this test (dangling-export detection)

🤖 Generated with Claude Code