Claude review skills fail silently when run via background Task subagent

[cderv]

Something I observed recently so sharing in case I am not the only one

---

The Claude skills for review commands instruct Claude Code to run the review using the Task tool with run_in_background: true and subagent_type: "Bash". In practice, the background agent completes but produces no usable output, and Claude Code reports it failed.

Reproduction

From a repo with roborev initialized:

/roborev-design-review-branch

Session output:

● Bash(Run branch design review)
  ⎿  Backgrounded agent
  ⎿  Prompt:
       Run this command in the directory C:\Users\chris\Documents\DEV_R\quarto-cli.worktrees\issue-13341:
       roborev review --branch --wait --type design

● Design review submitted for the branch. I'll present the results when it completes — you can keep working in the meantime.

● Agent "Run branch design review" completed

● The background agent wasn't able to run the command. Let me run it directly.

The agent completes and Claude interprets the result as a failure. The actual agent output isn't visible in the session, so the specific failure mode is unclear.

Affected skills

All four Claude review skills contain the same pattern at internal/skills/claude/*/SKILL.md:

Use the `Task` tool with `run_in_background: true` and `subagent_type: "Bash"`:

• roborev-review/SKILL.md (line 49)
• roborev-review-branch/SKILL.md (line 49)
• roborev-design-review/SKILL.md (line 48)
• roborev-design-review-branch/SKILL.md (line 48)

The Codex versions of these skills run the command synchronously and don't use the background Task pattern, so they may not be affected.

Possible causes

A few things could explain the failure:

• Background subagents in Claude Code auto-deny tool permissions that weren't pre-approved before launch. If the Bash permission pre-approval doesn't trigger for the roborev review command, it gets silently denied.
• The --wait flag blocks until the review completes, which could hit a timeout in the background agent before the review finishes.
• The command could run but return an error (daemon not running, wrong working directory) that the agent doesn't surface.

I don't have visibility into what the background agent actually returned, so I can't narrow it down further.

Ideas

• The --wait flag combined with a background subagent seems like it could cause timeout issues. The command blocks, the agent blocks, and if the review takes longer than the agent timeout, it fails. Removing --wait would let the enqueue return immediately, avoiding the timeout window.
• Claude Code's Bash tool supports run_in_background: true natively. Using Bash directly instead of wrapping it in a Task subagent would avoid the subagent permission model entirely — the user approves the command in the foreground, then it runs in the background.

Environment

• roborev at ebf5478 (main)
• Claude Code with Claude Opus 4.6
• Windows 11, though likely not Windows-specific