$ the-wire · showcase
Cline's GPT-6 fix ships across desktop, CLI and SDK; Codex tightens plugin hook trust
By RepoJournal · Filed · About Agentic Coding · Composed from the cited sources · methodology
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?v=4)
23 people shipped this
github-actions[bot]
@github-actions[bot]
2 cited
copyberry
@copyberry
1 cited
Benjamin Carlsson
@bc-openai
1 cited
vimtor
@vimtor
1 cited
GitHub Actions
@actions-user
Eric Traut
@etraut-openai
opencode-agent
@opencode-agent
neriousy
@neriousy
Aiden Cline
@rekram1-node
opencode-agent[bot]
@opencode-agent[bot]
villahernandez-coder
@villahernandez-coder
urielefrenvirtusa
@urielefrenvirtusa
jvargassanchez-dot
@jvargassanchez-dot
ugorla-dev
@ugorla-dev
Engel Nyst
@enyst
Vasco Schiavo
@VascoSch92
ak684
@ak684
abeatrix
@abeatrix
Mikołaj Kondratek
@mkondratek
jamadeo
@jamadeo
jbg
@jbg
Alex Hancock
@alexhancock
gokay-ai
@gokay-ai
Cline's GPT-6 requests now use max_completion_tokens, unblocking OpenAI and Azure endpoints across desktop v0.0.48, CLI v3.0.73 and SDK v0.0.95, while Codex binds plugin hook approvals to the plugin's actual contents rather than its declaration.
Desktop v0.0.48 cline/cline
Desktop v0.0.48 fixes the request bug that had OpenAI and Azure rejecting every GPT-6 call because Cline sent max_tokens instead of max_completion_tokens, and returns Claude Sonnet 5.5 to the default model slot for the Cline provider. The model catalog refresh adds CommandCode and Reflection as providers and changes defaults for 302.AI (Gemini 4 Argon), Eden AI and Prime Inference: if you pinne...
Bind plugin hook approval to plugin contents openai/codex
A plugin's scripts could change without its hook declaration changing, so declaration-only approval let modified plugin code run under an approval you already granted. Codex now fingerprints the plugin bundle and persists the content hash as trusted_content_hash, verifying contents before each command or MCP hook executes while keeping trusted_hash readable for older clients.
CLI v3.0.73 cline/cline
The CLI release carries the same GPT-6 max_completion_tokens fix and the same return of Claude Sonnet 5.5 as the no-model default, plus the catalog additions of CommandCode and Reflection. The same fix lands in SDK v0.0.95, where Azure's gpt-35-turbo keeps max_tokens and gpt-6-1-sol is included alongside gpt-6-sol and gpt-6.1-sol.
feat(core): read model reasoning from config anomalyco/opencode
OpenCode gains providers.*.models.*.capabilities.reasoning in models.dev v2's shape, so a model's reasoning controls can be set as none, fixed, effort levels, or a token budget instead of hand-writing raw request bodies per provider. Org admins set these in the OpenCode Console and they arrive in the pulled config.
Stop applying macOS background priority to managed daemons (#54763) openai/codex
Managed app-server and updater processes now launch through the shared background command helper without the macOS taskpolicy -b wrapper and its availability probe. The 30-second macOS readiness timeout stays, because older updaters can still pass background policy to replacement daemons during upgrades. Elsewhere it was a maintenance day: Gemini CLI fixed keybindings.json entries that use a ke...
Action items
