fix-issue-batch: agent cuts corners compared to standalone fix-issue

[isPANN]

Problem

When running /fix-issue-batch, the agent is less thorough than when running /fix-issue <N> directly on a single issue. Observed behaviors:

• Skips or superficially handles warnings (Warn results) that standalone /fix-issue would address
• Auto-fixes without properly presenting them to the human for review
• Rushes through substantive issues instead of brainstorming one-at-a-time with options
• May skip the subagent re-check step or not wait for proper human approval

Expected Behavior

Each issue processed by /fix-issue-batch should receive exactly the same level of scrutiny as a standalone /fix-issue call — the batch wrapper should only handle queue management (fetching, sorting, progress tracking), not alter the quality of per-issue processing.

Possible Causes

• The batch context accumulates too many tokens, causing the agent to compress/shortcut later issues
• The agent "pattern matches" on batch = fast and unconsciously reduces effort per item
• The fix-issue-batch skill delegates via /fix-issue but the agent may inline the logic instead of following the full skill steps

Potential Fixes

• Add explicit instruction in fix-issue-batch that each issue must follow all fix-issue steps without shortcuts
• Consider dispatching each /fix-issue call as a separate subagent so it gets a fresh context window and cannot be influenced by batch fatigue
• Add a "quality gate" check: after each issue, verify that the changelog comment was posted and all Fail/Warn items were addressed