fix(grok): reject unsafe protobuf response lengths

[luvs01]

Motivation

• The hand-written protobuf varint decoder used bitwise shifts that coerced results to signed 32-bit integers, allowing a crafted Grok response to produce a negative length and drive a synchronous CPU denial-of-service.
• The decoder and top-level parser trusted provider-controlled lengths and advanced offsets without validating non-negativity, safe-integer representation, or bounds inside the enclosing message.

Description

• Replace signed 32-bit bitwise accumulation in decodeVarint with safe-number arithmetic and validate the offset argument by throwing on invalid inputs.
• Reject truncated or unsafe varints by throwing on overflow/unsafe integers and on incomplete encodings instead of returning a corrupted value.
• Add a decodeLength helper that validates a length-delimited field (length is non-negative, a safe integer, and does not exceed the remaining bytes) and use it in decodeTimestamp, decodeConsumerResetToken, and decodeGetRemainingResetsResponse so parser offsets advance only within bounds.
• Add a regression test that asserts the reported malicious 2^31 length payload is rejected and that truncated varints throw, and document the src/grok/ parsing contract in structure/runtime.md.

Testing

• Ran bun test tests/providers/xai/grok-reset-coupons.test.ts, which passed (7 tests, focused suite).
• Ran bun run structure:check, bun run typecheck, and bun run privacy:scan, all of which passed.
• Attempted bun run test (full suite); the focused security regression passed, but the full parallel run encountered unrelated worker panics and admission-test failures in other domains, so the repository-wide run did not complete cleanly.
• Verified git diff --check and the local changes compile and the focused regression prevents the previously reported exploit input from causing synchronous infinite parsing.

---

Codex Task

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Advanced

Run ID: 8b4c5b26-f77c-46e3-b72a-1a92a9f495d6

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-14T00:24:10.610069Z	c2e5e77	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[github-actions]

✅ Deterministic PR hygiene checks passed.