chore(release): Bump version to 0.4.1

[arcaputo3]

Summary

• Bump build.mill fallback version to 0.4.1
• Bump package.json version to 0.4.1
• Update README.md version references to 0.4.1

What's in v0.4.1

• SDK bump to 0.2.75 (parity with Claude Code v2.1.75)
• AgentMcpServerSpec ADT for subagent MCP server configuration
• criticalSystemReminder_EXPERIMENTAL on AgentDefinition
• onElicitation callback on AgentOptions
• Session management: renameSession, tagSession, getSessionInfo
• SessionLogger hardening (JSON escaping, opt-in log dir)
• AnyOf schema support — JsonSchema.AnyOfType + anyOf() constructor for union type schemas, with Zod conversion via z.union()
• ToolInput summon for ADT enums — macro tries Expr.summon[ToolInput[T]] before auto-generating schemas, enabling custom JSON encoding for sealed ADT enums

Post-merge

1. Tag: git tag -a v0.4.1 -m "Release 0.4.1"
2. Push tag separately: git push origin v0.4.1
3. CI publishes to Maven Central
4. Bump to 0.4.2-SNAPSHOT for development

🤖 Generated with Claude Code

[claude]

PR Review: chore(release): Bump version to 0.4.1

This is a straightforward version bump PR. Here's my assessment:

Summary of Changes

• build.mill: Fallback publishVersion changed from 0.4.1-SNAPSHOT → 0.4.1
• package.json: version changed from 0.4.1-SNAPSHOT → 0.4.1
• README.md: All 5 version references updated from 0.4.0 → 0.4.1

Observations

Consistency note — README vs build.mill snapshot baseline

The README previously said 0.4.0 (last published release), while build.mill and package.json were already on 0.4.1-SNAPSHOT. This PR resolves that inconsistency correctly by locking everything to 0.4.1.

build.mill fallback version

The pattern Task.env.get("PUBLISH_VERSION").getOrElse("0.4.1") is fine for release, but it means anyone who builds locally without setting PUBLISH_VERSION will produce artifacts stamped 0.4.1. Once this merges and the release tag is cut, you'll want to immediately bump to 0.4.2-SNAPSHOT (as noted in the post-merge checklist) so local builds are distinguishable from the published release. The checklist already calls this out — just flagging it as the most important follow-up.

package.json version

package.json is marked "private": true, so its version field has no publishing impact. The bump is still good for consistency and changelog tooling.

No test changes needed

This is purely a metadata/version bump with no logic changes, so the absence of new tests is appropriate.

Verdict

✅ Looks good to merge. The changes are minimal, correct, and internally consistent. The post-merge checklist (tag → push tag → wait for CI publish → bump to 0.4.2-SNAPSHOT) is exactly the right process.

🤖 Reviewed with Claude Code

[claude]

Code Review: v0.4.1 Release PR

Overall this is a clean, well-scoped release PR. The version bumps are consistent across all three files (build.mill, package.json, README.md), and the two feature additions are focused and follow existing patterns well.

---

JsonSchema.AnyOfType + anyOf() (ToolDef.scala)

Looks good. The new AnyOfType case class and anyOf() constructor fit naturally into the existing JsonSchema ADT. toRaw is consistent with all other schema types.

One edge case to consider: Zod.union([]) with an empty list may throw at runtime. The anyOf() constructor accepts schemas: JsonSchema* with no minimum cardinality enforcement. A guard would prevent surprising errors:

// Optional: fail fast at construction time
def anyOf(schemas: JsonSchema*): JsonSchema =
  require(schemas.nonEmpty, "anyOf requires at least one schema")
  AnyOfType(schemas.toList)

---

ZodConverter (ZodFacade.scala)

The added case is correct and minimal. No issues.

---

trySummonToolInput macro (ToolInputMacros.scala)

The design is sound — summon-first is the right priority for custom overrides. The implementation is clean and the comment explaining the BodyContent use case is helpful.

One observation: The summon priority currently only applies to Flags.Enum types and the other (unknown) fallback — it does not apply to case classes (Flags.Case). This means a user cannot override the auto-derived schema for a case class with a custom ToolInput[MyCaseClass]. This may be intentional, but it's worth documenting as a known limitation if so. If override capability is desired for case classes too, the same trySummonToolInput pattern could be added there.

Potential macro footgun: Expr.summon[ToolInput[t]] silently returns None if no instance is in scope. If a user defines a ToolInput given in a file that isn't imported at the macro call site, they'll get the auto-generated schema with no warning. This is standard Scala implicit behavior, but worth noting in the docs/usage comment.

---

Test coverage

Neither AnyOfType/anyOf() nor the trySummonToolInput path have tests in this PR. For a release, it would be good to have at minimum:

• A unit test for anyOf toRaw serialization (e.g., anyOf(JsonSchema.string, JsonSchema.array(JsonSchema.string)).toRaw produces {"anyOf": [...]})
• A macro expansion test for a type with a custom given ToolInput[T] to confirm the summon path is hit

These could be follow-ups if test infrastructure for macros is non-trivial to set up, but the toRaw serialization test would be straightforward with bun test.

---

Version bump hygiene

The build.mill fallback default changing from 0.4.1-SNAPSHOT to 0.4.1 is correct and expected. The post-merge checklist in the PR description (tag → push → CI publish → bump to 0.4.2-SNAPSHOT) is clear.

---

Summary: Approve with the suggestion to add serialization tests for AnyOfType before or shortly after merging, and optionally add an input guard on anyOf() for empty schemas.

[claude]

PR Review: chore(release): Bump version to 0.4.1

Overall this is a clean release PR. The version bumps are straightforward and the two substantive changes (macro improvement + AnyOf schema support) are well-structured. A few observations below.

---

✅ What's good

• trySummonToolInput pattern is a solid design — checking for explicit ToolInput[T] instances before falling back to auto-generation is the right priority order. It makes the macro extensible without breaking existing users.
• AnyOfType follows the existing JsonSchema ADT pattern cleanly, and the Zod conversion (Zod.union) maps correctly.
• The trySummonToolInput helper is properly shared across all three macro branches (enum, case class, unknown), which is consistent.

---

⚠️ Potential Issues

1. Zod union requires ≥ 2 types at runtime

Zod throws if z.union([]) or z.union([oneType]) is called. The anyOf() constructor has no guard:

def anyOf(schemas: JsonSchema*): JsonSchema = AnyOfType(schemas.toList)

Consider validating at construction time:

def anyOf(schemas: JsonSchema*): JsonSchema =
  require(schemas.length >= 2, s"anyOf requires at least 2 schemas, got ${schemas.length}")
  AnyOfType(schemas.toList)

2. AnyOfType with a single schema in ZodConverter

Similarly, if AnyOfType(List(one)) somehow reaches ZodConverter.toZodType, it will produce Zod.union(js.Array(oneType)) which will throw at runtime. A defensive fallback like unwrapping single-schema AnyOf to the schema itself would be safer.

3. Potential implicit summon ambiguity for case classes

The Case branch now calls trySummonToolInput first. If the macro-derived given ToolInput[T] is in scope at the call site (from a derives ToolInput on the same type), this could cause unexpected behavior — the summon might either succeed with the derived instance (circular) or fail, depending on Scala's implicit resolution order during macro expansion. Worth a note in the docs or a targeted test case.

---

🧪 Test Coverage

There are no tests added for:

• JsonSchema.anyOf(...) / AnyOfType.toRaw output
• ZodConverter handling of AnyOfType
• The trySummonToolInput path for enums and case classes with explicit ToolInput instances

Given these are user-facing API additions, even a few unit tests in the MUnit suite would help prevent regressions.

---

📋 Minor

• The fallback version in build.mill changing from 0.4.1-SNAPSHOT → 0.4.1 is correct for a release PR. The post-merge note about bumping to 0.4.2-SNAPSHOT is a good reminder — consider whether that bump should be a follow-up PR rather than a manual step.
• package.json version bump is consistent. Since this is a Scala library and package.json is marked "private": true, this is mainly for tracking — looks fine.

---

Summary: The core implementation is sound. The main risk is the missing lower-bound validation on anyOf / AnyOfType that could surface as a runtime Zod error. Recommend adding that guard plus at minimum a couple of regression tests before merging.

🤖 Generated with Claude Code