fix(rivetkit): bind methods through createWriteThroughProxy

[abcxff]

Fixes an error that occurs when inspector state attempts to serialize non trivial values (ie Date).

[railway-app]

🚅 Deployed to the rivet-pr-4987 environment in rivet-frontend

Service	Status	Web	Updated (UTC)
website	😴 Sleeping (View Logs)	Web	May 12, 2026 at 2:57 pm
frontend-inspector	❌ Build Failed (View Logs)	Web	May 12, 2026 at 2:48 pm
frontend-cloud	❌ Build Failed (View Logs)	Web	May 12, 2026 at 2:47 pm
ladle	❌ Build Failed (View Logs)	Web	May 12, 2026 at 2:46 pm
kitchen-sink	❌ Build Failed (View Logs)	Web	May 12, 2026 at 2:45 pm
mcp-hub	✅ Success (View Logs)	Web	May 6, 2026 at 8:30 pm

[abcxff]

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

• DO NOT MERGE: queue diagnose #5050
• fix(rivetkit): keep actor alive while workflow waits for queue messages #5052
• [slop]fix(workflow-engine): only commit step state after success #5010
• temp: fix wasm-pack #5048
• fix(rivetkit): bind methods through createWriteThroughProxy #4987 👈 (View in Graphite)
• feat(rivetkit): add ConnectionMap readonly Map wrapper for actor connections #5047
• fix(rivetkit-core): chunk apply_batch puts/deletes at 128 #5046
• fix(pegboard): validate drain_grace_period < request_lifespan #4998
• chore(rivetkit): log inspector errors + standardize pino errorKey error #4986
• fix(rivetkit): remove async drizzle db proxy #4984 : 2 other dependent PRs (#4985 , #4991 )
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[claude]

Code Review: PR #4987 -- fix(rivetkit): bind methods through createWriteThroughProxy

Overview

This PR fixes a serialization bug where inspector state fails to serialize non-trivial values like Date. The root cause: the old hand-rolled Proxy-based createWriteThroughProxy only trapped set and deleteProperty, so mutations via methods (e.g. date.setFullYear(), map.set(), set.add()) were never detected — those go through a get trap to retrieve the method followed by a call, bypassing the set trap entirely.

The fix replaces the hand-rolled proxy with @rivetkit/on-change, which correctly intercepts mutations on Date, Map, Set, TypedArrays, and arrays via method calls. It also introduces a CborSerializable type, assertCborSerializable validation, and adds Set round-trip support to the JSON-compat encoding layer.

The approach is sound and the test coverage is comprehensive.

---

Issues

1. RegExp was previously allowed but is now rejected

The old isCborSerializable returned true for RegExp (citing CBOR tag 27). The new assertCborSerializable throws for it. This is a breaking state-schema change for any user with RegExp in actor state. If intentional, it should be noted in the PR description.

2. Non-finite number validation regression

The old isCborSerializable rejected NaN and Infinity via !Number.isFinite(value). The new assertCborSerializable passes all typeof value === 'number' values through unconditionally. Encoding NaN will produce output but decoding may silently corrupt state. Recommend adding the guard back:

if (typeof value === 'number') {
    if (!Number.isFinite(value)) {
        throw new TypeError(
            `Value at ${path || 'root'} is non-finite (NaN/Infinity) and is not CBOR serializable`
        );
    }
    return;
}

3. $Set backward-compatibility risk

Any existing serialized actor state shaped like a two-element array starting with "$Set" will now be incorrectly deserialized as a Set instance after this change (same pattern risk as the existing $BigInt/$ArrayBuffer tags). Probably very unlikely in practice, but worth noting.

4. onValidate always returns true

onValidate(_path: string, newValue: unknown) {
    beforeChange?.(newValue);
    return true;
},

The unconditional return true is easy to misread. Rejection works by beforeChange throwing (the test confirms this). A short comment clarifying that rejection is throw-based, not return-value-based, would help future readers.

---

Minor Notes

• @rivetkit/on-change is Rivet-scoped. Worth a note in the PR clarifying whether this is a Rivet-maintained fork or the upstream on-change package re-scoped.
• Pinned version "@rivetkit/on-change": "6.0.1" (no caret) is a reasonable stability choice.
• The as CborSerializable casts throughout are appropriate since real validation happens at runtime via assertCborSerializable at write paths, and call sites typically hold unknown.
• Replacing isCborSerializable (boolean check) with assertCborSerializable (throw-on-invalid) aligns with the fail-by-default convention in CLAUDE.md.
• Test coverage is thorough: all CBOR types, deep nesting, Map/Set/Date/TypedArray method mutations, beforeChange throwing, and Set encoding round-trips. Well done.

---

Summary

The core fix is correct and well-motivated. The two behavioral regressions (non-finite numbers, RegExp) should be addressed before merge — particularly the NaN/Infinity gap, which is a silent correctness issue rather than a noisy failure.