fix(cli): use full GPT-5.6 OAuth context

[marius-kilocode]

Kilo currently reports a 500k context limit for ChatGPT OAuth GPT-5.6 models, even though the current Codex limits support a 1.05M-token context window and 922k input window. This causes Kilo to compact or reject conversations earlier than necessary.

Use the current Codex context and input limits for every GPT-5.6 OAuth model while preserving the existing GPT-5.5 behavior. The release metadata documents the user-facing limit increase.

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

Files Reviewed (3 files)

• .changeset/gpt-56-oauth-context.md
• packages/opencode/src/plugin/openai/codex.ts
• packages/opencode/test/plugin/codex.test.ts

The change is a minimal, well-scoped value update: GPT-5.6 OAuth models now use the current Codex limits (1.05M context / 922k input), the kilocode_change markers are consistent with the existing inline pattern in this shared file, the updated test assertions match the new plugin values, and the changeset is user-facing and correctly scoped as a patch. No stale references to the old 500k/372k limits exist elsewhere in the repo, and no resource-lifecycle or memory-leak concerns are introduced.

---

_{Reviewed by kimi-k3 · Input: 65K · Output: 4.5K · Cached: 200.7K}

_{Review guidance: REVIEW.md from base branch main}