fix: capture and display ai-reviewer.sh error messages in workflow

[LearningCircuit]

Summary

• Fixes workflow error handling to display actual error messages from ai-reviewer.sh
• Previously, failures would show only "Error: Process completed with exit code 1" without details
• Now captures and displays the full error output for easier debugging

Changes

• Temporarily disable shell errexit before running ai-reviewer.sh
• Capture both stdout and stderr with proper redirection
• Store and check exit code separately
• Display full error output when script fails
• Include exit code in debug mode output

Test plan

• The next AI review run will properly display any error messages if failures occur
• This PR itself will test the error handling when the AI review is triggered

Related

• Addresses the issue seen in feat: integrate document processing into NewsScheduler system local-deep-research#929 where error messages were hidden

[github-actions]

AI response could not be processed. Please check the workflow logs for debugging information.

[github-actions]

AI Code Review

This PR effectively fixes error handling and debugging output contamination issues in the AI reviewer workflow.

🔒 Security

• No security concerns identified. The 2>&1 stderr capture is safe here as the script only outputs progress messages (now gated by DEBUG_MODE). No sensitive data exposure risk from the changes.

Code Quality & Best Practices

✅ Strengths:

• Proper error capture pattern: disabling errexit, capturing exit code, then re-enabling
• DEBUG_MODE gating prevents stderr contamination of JSON output, fixing the parsing issue mentioned in commit history
• Clear error messaging with exit code and full output for debugging
• Graceful degradation when GitHub API calls fail (intentional design)

⚠️ Minor Improvements:

• Useless use of cat: cat diff.txt | bash ai-reviewer.sh should be bash ai-reviewer.sh < diff.txt for efficiency
• Overly aggressive error suppression: gh api ... 2>/dev/null || echo "" hides authentication/network issues. Consider logging API failures to stderr when DEBUG_MODE=true
• DRY principle: The repeated if [ "$DEBUG_MODE" = "true" ] checks could be refactored into a helper function, though this is acceptable for shell scripts

Performance

• No impact. Changes are purely control-flow and output-related.

Testing Considerations

• The fix itself will validate the error handling on this PR's AI review run
• Consider testing the failure path by temporarily introducing a syntax error in ai-reviewer.sh

---

Review by Friendly AI Reviewer - made with ❤️