[BUG] Scheduled routines silently no-op when the target repo checkout is missing — no dry-run/precondition check

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

A scheduled/cron Claude Code session was configured against a specific repo with an elaborate multi-step review prompt. When it fired, the session's working directory was empty and not a git repository — the repo had never been checked out into the environment. There was no gh CLI, no configured remote, and no clone anywhere on disk to recover from.

The session correctly detected this and sent a failure notification instead of fabricating a review, but the deeper problem is that this is discoverable at schedule-creation time, not just at run time. (A human using this feature is apt to assume that Claude has cloned the repo into the sandbox it'll be using; after all it knows how to commit.) A scheduled task tied to a repo should be validated (or dry-run) when the schedule is created/edited so a broken binding surfaces immediately, rather than silently failing on whatever day it happens to fire next — potentially days or weeks after setup, with the user unaware anything was ever wrong until a "your routine failed" notification (or worse, no notification at all if the failure mode is different).

What Should Happen?

Add a dry-run/precondition check when a scheduled/cron task is created or edited (e.g., verify repo checkout succeeds, working directory matches expectations) so misconfiguration is caught at setup time, not discovered N days later.

Error Messages/Logs

Steps to Reproduce

1. Create a recurring task to perform an AI-centric operation against a repo (e.g. review code and tests)
2. Neglect to configure the web-based Claude with a copy of the repo on the assumption that it already knows how to access the repo

Claude Model

None

Is this a regression?

No, this never worked

Last Working Version

_No response_

Claude Code Version

whatever Claude.ai/code is running as of 2026-07-27 8:00AM Pacific (Mountain View) time

Platform

Anthropic API

Operating System

Other Linux

Terminal/Shell

Other

Additional Information

I configured this using the web UI (Claude.ai/code)