feat: add "Auto" runtime mode — AI-reviewed approvals for Codex and Claude

[t3dotgg]

Summary

Adds a new "auto" runtime mode between Auto-accept edits and Full access. Both native backends recently shipped a model-reviewed approval mode that is strictly safer than full access — risky actions can still be denied or escalated, and (for Codex) the sandbox stays on:

• Codex: keeps approvalPolicy: "on-request" + sandbox: workspace-write, and sets approvalsReviewer: "auto_review" on thread/start and turn/start — a reviewer subagent decides approval requests (sandbox escapes, network access, MCP prompts) instead of routing them to the user.
• Claude: passes permissionMode: "auto" (no allowDangerouslySkipPermissions), where a classifier model approves/denies each permission prompt. Escalations still flow through canUseTool into the existing approval UI; classifier denials surface via the already-handled permission_denied → tool.denied event.
• Other adapters (Cursor, Grok, OpenCode): fall through to their existing non-full-access behavior.

Changes

• packages/contracts: add "auto" to RuntimeMode
• apps/server: Codex + Claude adapter mappings
• apps/web: mode picker entries (full + compact composer)
• apps/mobile: runtime options in ThreadComposer and NewTaskDraftScreen
• Tests for both adapters' wire params

Notes for reviewers

• No DB migration needed: runtime_mode is a TEXT column with no CHECK constraint; mid-thread mode changes already restart the session via ProviderCommandReactor.
• approvalsReviewer exists in the generated app-server schema (effect-codex-app-server), so no regeneration was needed. Requires a codex binary recent enough to know the field (verified in 0.145.0).
• Older mobile clients pinned to the previous contract will fail to decode threads set to "auto" — same skew consideration as any enum addition.

🤖 Generated with Claude Code

---

Note

Medium Risk
Changes approval routing and sandbox/permission semantics for agent sessions; incorrect mapping could weaken supervision or strand users on sticky reviewers after mode changes.

Overview
Adds auto as a new RuntimeMode between auto-accept edits and full access, exposed in web, mobile, and compact composer pickers.

Codex maps auto to on-request approvals with workspace-write sandbox and sets approvalsReviewer: auto_review on every thread/turn start (including non-auto modes with explicit user) so mode switches do not leave the auto reviewer sticky on resume.

Claude maps auto to permissionMode: "auto" without allowDangerouslySkipPermissions.

Adapter tests cover the new wire params for both backends.

^{Reviewed by Cursor Bugbot for commit 8c2ff69. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add 'Auto' runtime mode with AI-reviewed approvals for Codex and Claude

• Adds 'auto' as a valid RuntimeMode in orchestration.ts, wiring it through the full stack: web/mobile UI menus, Claude adapter, and Codex session runtime.
• In Claude, runtimeMode: 'auto' maps to permissionMode: 'auto' in outgoing query options via ClaudeAdapter.ts.
• In Codex, 'auto' mode sets approvalsReviewer: 'auto_review', approvalPolicy: 'on-request', and uses the workspaceWrite sandbox policy via CodexSessionRuntime.ts.
• All other runtime modes now explicitly set approvalsReviewer: 'user' in thread and turn start params.

^{Macroscope summarized 8c2ff69.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro

Run ID: 46401bd4-c7a4-4f37-95d0-8a591a77f040

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch t3code/plan-auto-approve-permissions

---

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

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit b742293. Configure here.}

[macroscopeapp]

Approvability

Verdict: Needs human review

This PR introduces a new user-facing feature ('Auto' runtime mode) that routes approval decisions to an AI reviewer instead of the user. New features with significant workflow changes warrant human review.

^{You can customize Macroscope's approvability policy. Learn more.}