$ the-wire · showcase
SDK RELEASES LOCK IN PROMPT CACHING AND SPEND CONTROLS
By RepoJournal · Filed · About OpenAI · Composed from the cited sources · methodology
Both Node and Python clients shipped v6.49.0 and v2.48.0 overnight with admin spend limit APIs and relaxed prompt cache validation.
OpenAI's Node SDK [6] and Python SDK [7] both released in sync, adding support for `spend_limit` admin APIs and accepting `None` values for `prompt_cache_key` and `safety_identifier` fields. The spend controls represent a significant shift in how teams manage API costs at scale, while the cache key relaxation unblocks workflows that need to explicitly clear cache state. On the Codex side, copyberry merged five major PRs addressing the plugin execution pipeline: trusted script attribution now tracks which plugin executed commands [1], resolves scripts against verified plugin roots [2], and leaves ambiguous paths unattributed for safety. Thread replay got incremental update support [4] so callers can fetch changed items without replaying the entire history, and remote skill icons now surface through the app server [5]. The installed app duration metrics also aligned with legacy baselines [3] to enable direct performance comparison. Across 3 repos, the period logged 50 commits, 48 PRs, and 5 releases.
Action items
- → Pin Node SDK to 6.49.0 and update spend limit calls in admin panels openai/openai-node [plan]
- → Update Python client to 2.48.0 and test prompt cache None handling openai/openai-python [plan]
- → Review plugin script execution logs for unattributed command paths openai/codex [monitor]
References
- [1] Attribute command executions to trusted plugin scripts ↗ openai/codex
- [2] Add trusted plugin script attribution ↗ openai/codex
- [3] Align installed app duration metrics with the legacy baseline ↗ openai/codex
- [4] Support incremental replay of updated thread items ↗ openai/codex
- [5] Expose remote skill icon URLs through app server ↗ openai/codex
- [6] v6.49.0 ↗ openai/openai-node
- [7] v2.48.0 ↗ openai/openai-python