[kitlangton]

Schedules can now run a Code Mode JavaScript check (if: { code } or until: { code }) before waking a session, using the same state and unattended lifecycle as shell checks, with no model call and no allow list to maintain.
Builds on #60137 (merged). Everything stays behind experimental.schedules, and requires no new database migration.

💡 Why
Shell checks cover local CLIs, while integrations like Slack, Linear, and MCP servers hold OAuth credentials inside OpenCode and often want structured JSON cursor state between checks.

🆕 Now you can

• Pass { code: "..." } to if or until on schedule_create or /api/experimental/schedule. OpenCode walks the Acorn AST at creation (CodeMode.scanTools) to find every tools.<path>(...) call statically, rejects dynamic tools[x] or aliasing, authorizes those tools once without saving a project-wide grant, and exposes only those tools (with no fetch) on unattended ticks.
• Read state, now and runs directly as globals (seeded at creation on every and until so events before the first tick aren't missed), and return either a boolean (until) or { wake, state?, message? }.
• Keep recursive or file-editing tools (opencode.schedule_*, session_move, worktree_*, and edit tools) out of unattended checks automatically.

🚦 How it ships
Off by default behind experimental.schedules, same as #59401, #59457 and #60137.

  schedule_create({ when, if / until: { code }, prompt })
+   -> CodeMode.scanTools(code) extracts static `tools.<path>(...)` calls from the AST
+   -> authorize extracted tools once (no `save`), seed initial `state` at t = 0
+   -> on each tick: run confined Code Mode over only extracted tools (no `fetch`, no model call)
    -> wake only when `wake: true`, appending `message` to `prompt`

📊 Proof: 4 new AST scanner and binding tests in @opencode/codemode, 46 core schedule/tool/codemode tests (covering t=0 state seeding, wake-only ends.times, boolean and state-free until returns, fetch absence, static tool extraction, one-time approval, denied secondary permissions, and forbidden recursive/edit tools), and 1 end-to-end server API test.

👀 Read: packages/codemode/src/scan.ts (scanTools) and authorizeCode / runCode in packages/core/src/schedule/check.ts. Tool.Service stays free of Permission.Service: ScheduleCheck owns the Permission.Unattended scope, while Tool.check only filters the inventory. The rest is routine (schema, OpenAPI, generated client).

_{Planned feature · no user reports · not measured}

Checks run

• bun test test/scan.test.ts (codemode): 4 pass, 0 fail
• bun run script/test.ts test/schedule.test.ts test/tool-schedule.test.ts test/codemode.test.ts (core): 46 pass, 0 fail
• bun test test/schedule.test.ts (server): 1 pass, 0 fail
• bun run check (full repo lint, sdk-docs, and 35-package turbo typecheck): pass