v0.3.3

v0.3.3 — Mid-stream fallback + gpt-5.6 cache keep-warm

• Reroute when an account hits its limit mid-generation (WebSocket). If the active account's quota is exhausted partway through a response, the request now fails over to a healthy fallback account on the same turn, instead of failing on the exhausted one. This applies before any output has streamed; once a response has started, the turn completes and the account is avoided on the next turn (so nothing is billed or run twice).
• gpt-5.6 prompt-cache keep-warm. Keep-warm now matches gpt-5.6's ~30-minute prompt cache (vs ~5 minutes on older models), with a bounded warm cap for short-lived subagent sessions and an optional /openai-cachekeep HH-HH clock window to limit warming to set hours.

v0.3.2

v0.3.2 — GPT-5.6 Pro reasoning + model preservation

• -pro models now use maximum reasoning effort. Selecting a -pro GPT-5.6 model (e.g. gpt-5.6-sol-pro) previously behaved identically to the base model — OpenCode expresses Pro as a reasoning mode the Codex backend does not accept, and it was dropped before the wire. The plugin now translates -pro to reasoning.effort: "max" (the Codex Pro tier), on the HTTP, WebSocket, and raw-WebSocket paths alike.
• Slash commands no longer change your model. Running an /openai-* command sent a hidden message that could cause OpenCode's next prompt to inherit a default model/variant, silently switching the active model. Command replies now carry the previous model, agent, and variant.

v0.3.1

This release resolves findings from a full security and correctness audit of the plugin. No user action required; upgrade in place.

Security

• Release CI no longer interpolates untrusted input into shell and pins all GitHub Actions to commit SHAs, closing a workflow injection / supply-chain path.
• Diagnostic dumps and log files are now written with private permissions (0600/0700), and credentials are redacted from dumped request bodies.
• main is now a reserved account id and cannot be used as a fallback label; the OAuth callback no longer discards a pending login on a code-less request.

Reliability

• The account killswitch no longer fails open on the WebSocket transport after a main-account re-login.
• A failed fallback account now records its quota, so the killswitch won't immediately retry an exhausted account; quota windows past their reset time are no longer treated as exhausted.
• Account fallback is restricted to the Codex responses endpoint and no longer retries after an indeterminate transport error, preventing duplicate requests.
• A crashed token refresh can no longer stall subsequent refreshes, and a rotated token is no longer lost if persistence briefly fails.
• First-run multi-account migration is serialized with account edits, and removed accounts are no longer refreshed.
• Fixed a file-lock recovery race that could surface under heavy concurrent load.

Internal

• HTTP request bodies are key-ordered to match the WebSocket path, avoiding a prompt-cache miss on transport fallback.
• Added CI test coverage for the Pi package.

v0.3.0

Full Changelog: v0.2.1...v0.3.0

v0.2.1

What's Changed

• test(accounts): assert removed-account secrets are pruned from the state file by @iceteaSA in #18
• fix(pkg): ship src/logger.ts and src/util for the raw-source ./tui entry by @iceteaSA in #8

Full Changelog: v0.2.0...v0.2.1

v0.2.0

Full Changelog: v0.1.4...v0.2.0

v0.1.4

What's Changed

• build(deps-dev): bump @opencode-ai/plugin from 1.17.1 to 1.17.8 by @dependabot[bot] in #6

New Contributors

• @dependabot[bot] made their first contribution in #6

Full Changelog: v0.1.3...v0.1.4

v0.1.3

What's Changed

• feat: ChatGPT multi-account parity (fallback, quota, cachekeep, modals, CLI) by @iceteaSA in #7

New Contributors

• @iceteaSA made their first contribution in #7

Full Changelog: v0.1.2...v0.1.3

v0.1.2

Full Changelog: v0.1.1...v0.1.2

v0.1.1

Full Changelog: https://github.com/cortexkit/openai-auth/commits/v0.1.1