[ergofobe]

Pin move to the v0.5.1 tag plus the version bump, in one PR because both are release mechanics for the same patch.

Why v0.5.1 of the SDK exists

The wire-boolean fix landed on SDK main as cb814ad, and #102 pinned the server there — legal between releases. But a consumer's release must name an SDK release, so shipping this meant cutting imogen-sdk v0.5.1 first and moving the pin onto the tag. That SDK release carries nothing but the fix.

git submodule status now reads imogen-sdk (v0.5.1). The only difference between the tag and the main commit it replaces is version files — verified by diffing them.

What this ships to users

Two live correctness defects in v0.6.0, both wire-boolean truthiness, both measured rather than inferred:

?includeHidden=false was read as true (#100, fixed in #101). Every SDK port defaults includeHidden to false and sends it, so an ordinary people.list() carried ?includeHidden=false — which arrived as true and went straight to listPeople. Hiding a person never hid them. That schema is declared here rather than in @imogen/shared, so it needed a server-side fix.

The same defect in the SDK's own schemas (#102) meant:

• ?trashed=false returned only trashed assets
• ?favorite=false returned only favourites
• an upload carrying favorite=false was favourited

Reachable from the shipped Rust client. That half could not be fixed here — redeclaring the SDK's fields is what CLAUDE.md forbids — so moving the pin is what actually delivers it.

Ranges and lockfile

The ^0.5.0 ranges already admit 0.5.1, so no manifest range moves. bun.lock needs no regeneration — confirmed before and after the bump, diff empty both times.

Verification

930 pass
0 fail
2021 expect() calls
Ran 930 tests across 48 files.

plus bun run --filter '@imogen/web' build exit 0, run both at the pin move and again after the version bump.

sdk-contract.test.ts — the only place the SDK and server meet — passes against the tag.

Review

Exempt: the pin move's substance was reviewed in #102, including an independent confirming review that re-derived its claims rather than accepting them (it verified the ?trashed=false chain through four independent links, swept the generated OpenAPI document for anyOf regressions, and overruled two reviewers on a .meta() claim it could not reproduce). This PR adds the tag move and version strings on top of that.

🤖 Generated with Claude Code