OpenCode Review

OpenCode Review

Iterative dual-review pipeline using both OpenCode (MiniMax M2.5) and Claude. Runs reviews, fixes issues, and repeats until both reviewers are satisfied.

Usage

/​opencode_review

Behavior

1. Run OpenCode Review

~/​.opencode/​bin/​opencode run -m opencode/​minimax-m2.5-free "/​review" 2>&1

Capture the full output. Parse it for:

• Specific file and line references
• Issue severity (bugs, style, security, performance)
• Actionable suggestions vs informational comments

If opencode is not found at ~/​.opencode/​bin/​opencode, tell the user to install it (curl -fsSL https://raw.githubusercontent.com/​opencode-ai/​opencode/​refs/​heads/​main/​install | bash) or add it to PATH.

2. Run Claude Self-Review

Review the same uncommitted changes yourself:

git diff
git diff --cached

Check for:

• Logic errors and bugs in the changed code
• Security issues (injection, exposed secrets, missing validation)
• Consistency with existing code patterns (read surrounding code if needed)
• Missing error handling at system boundaries
• Test coverage gaps for new/​changed functionality

3. Consolidate Issues

Merge findings from both reviews. Deduplicate — if OpenCode and Claude flag the same issue, count it once. Categorize:

• Must fix: Bugs, security issues, logic errors
• Should fix: Style violations, missing edge cases, unclear code
• Informational: Suggestions, minor nits, opinions

4. Fix Must-Fix and Should-Fix Issues

For each actionable issue:

1. Read the file and understand the context
2. Apply the fix
3. Keep fixes minimal — only address the flagged issue

Do NOT fix informational/​nit issues unless trivially adjacent to a real fix.

5. Re-Run Both Reviews

After fixing, run both reviewers again:

~/​.opencode/​bin/​opencode run -m opencode/​minimax-m2.5-free "/​review" 2>&1

And repeat your own Claude review of the current diff.

6. Iterate

If new issues are found:

• Fix them and re-run (back to step 5)
• Keep iterating until both reviewers are clean
• If you're unsure how to fix an issue or uncertain about the intended behavior, ask the user before guessing

If both reviews come back clean: proceed to summary.

7. Summary

## OpenCode Review Results

### Iteration 1
**OpenCode found:** 3 issues (1 bug, 2 style)
**Claude found:** 2 issues (1 logic error, 1 missing null check)
**Deduplicated total:** 4 issues
**Fixed:** 4/​4

### Iteration 2
**OpenCode found:** 0 issues
**Claude found:** 0 issues

### Status: CLEAN

### Fixes Applied
1. `src/​api/​tournaments.ts:45` — Added null check for optional `season` field (bug)
2. `src/​components/​Bracket.tsx:112` — Fixed off-by-one in round calculation (logic error)
3. `src/​utils/​scoring.ts:30` — Consistent naming: `calcScore` → `calculateScore` (style)
4. `src/​api/​tournaments.ts:78` — Added input validation on `status` parameter (missing check)

### Remaining (informational only)
- OpenCode suggested extracting lines 45-60 into a helper — skipped (one-time operation)

Ready to commit.

Important Rules

• Keep iterating until both reviewers are clean. If stuck on an issue, ask the user.
• Fix source code, not symptoms. If OpenCode flags a symptom, trace to the root cause.
• Don't fix nits in a loop. Informational items are reported but not acted on — they don't trigger another iteration.
• Both reviewers must be clean to declare success. One clean + one with issues = keep iterating.
• Don't argue with OpenCode. If it flags something legitimate, fix it. If it flags something wrong, note it in the summary as a false positive and move on.
• Never suppress warnings by adding ignore comments (e.g., // eslint-disable, @ts-ignore) unless the user explicitly asked.