Snapshot can recursively ingest its own data when worktree is the home directory #20442

Description

On OpenCode 1.2.24, I hit a disk exhaustion / high-CPU loop caused by the snapshot mechanism when the worktree was the home directory.

In my case, starting opencode serve from /home/<user> resulted in a snapshot repo under ~/.local/share/opencode/snapshot/global, and I observed repeated commands of the form:

git --git-dir /home/<user>/.local/share/opencode/snapshot/global --work-tree /home/<user> add

The problem is that the snapshot storage directory itself lives inside that same worktree. On this setup, ~/.local/share/opencode/ was visible to the snapshot repo as content under the captured tree instead of being reliably excluded. After process restarts / interrupted snapshot writes, objects/pack accumulated many very large tmp_pack_* files until the disk filled up, while git add kept respawning at high CPU.

This looks like a logic hole around broad worktrees such as /home/<user>: the snapshot repo can end up recursively ingesting its own data.

Expected behavior: snapshot storage should never be snapshotted. OpenCode should either store snapshot data outside the worktree, hard-exclude its own data directory, or reject worktrees that contain the OpenCode data directory.

Plugins

None

OpenCode version

1.2.24

Steps to reproduce

1. Start opencode serve with a worktree that contains the OpenCode data directory, for example /home/<user>.
2. Let OpenCode initialize snapshot storage under ~/.local/share/opencode/snapshot/....
3. Restart or interrupt the process while snapshot activity is happening.
4. Check ~/.local/share/opencode/snapshot/.../objects/pack.
5. Observe large tmp_pack_* files accumulating, rising disk usage, and repeated git add activity.

Screenshot and/or share link

No share link. I can provide additional logs if needed.

Operating System

Debian GNU/Linux 13

Terminal

Headless server environment