Summary

PlannerWorktreeStrategy._get_base_sources (git_worktree.py) runs sequential git show calls for each changed file. With max_files=50+, this adds measurable latency.

Finding source

Pre-PR review agents (python-reviewer, async-concurrency-reviewer) flagged this during #611 review.

Proposed fix

Use asyncio.TaskGroup with a bounded semaphore to parallelize the git show calls. The method runs under self._lock so no race conditions with other operations.

async with asyncio.TaskGroup() as tg:
    for file_path in files:
        tg.create_task(_fetch(file_path))

Context

• CLAUDE.md: "prefer asyncio.TaskGroup for fan-out/fan-in parallel operations in new code"
• Same optimization applies to CompositeSemanticAnalyzer running analyzers sequentially