From bee-default
External code review via OpenCode CLI - sends diff to a second AI model (e.g., Qwen3-Max, GPT-5-Nano) for independent review. Use after Bee's 6 parallel reviewers complete, as a complementary external perspective. Integrates with bee:requesting-code-review pipeline.
npx claudepluginhub luanrodrigues/ia-frmwrk --plugin bee-defaultThis skill uses the workspace's default tool permissions.
Uses [OpenCode CLI](https://github.com/opencode-ai/opencode) to send code diffs to an external AI model for independent review. This provides a **second AI perspective** from a different model, catching issues that Bee's internal reviewers might miss due to shared model biases.
Generates design tokens/docs from CSS/Tailwind/styled-components codebases, audits visual consistency across 10 dimensions, detects AI slop in UI.
Records polished WebM UI demo videos of web apps using Playwright with cursor overlay, natural pacing, and three-phase scripting. Activates for demo, walkthrough, screen recording, or tutorial requests.
Delivers idiomatic Kotlin patterns for null safety, immutability, sealed classes, coroutines, Flows, extensions, DSL builders, and Gradle DSL. Use when writing, reviewing, refactoring, or designing Kotlin code.
Uses OpenCode CLI to send code diffs to an external AI model for independent review. This provides a second AI perspective from a different model, catching issues that Bee's internal reviewers might miss due to shared model biases.
OpenCode CLI must be installed and configured:
which opencode || echo "NOT INSTALLED"
If not installed, prompt user to install or skip.
which opencode
| Result | Action |
|---|---|
| Found | Proceed to Step 2 |
| Not found | Ask user: install OpenCode or skip review |
Gather the code changes to review:
# Auto-detect base branch
BASE_SHA=$(git merge-base HEAD main 2>/dev/null || git merge-base HEAD master 2>/dev/null)
# Get the diff
git diff "$BASE_SHA"..HEAD > /tmp/bee-opencode-review-diff.txt
# Get changed files list
git diff --name-only "$BASE_SHA"..HEAD > /tmp/bee-opencode-review-files.txt
# Get commit messages for context
git log --oneline "$BASE_SHA"..HEAD > /tmp/bee-opencode-review-commits.txt
Use the comprehensive review instructions from docs/opencode-instructions.md.
The instructions file covers:
Run OpenCode with the review instructions and diff as file attachments:
opencode run \
-m "bailian-coding-plan/qwen3-max-2026-01-23" \
-f docs/opencode-instructions.md \
-f /tmp/bee-opencode-review-diff.txt \
--format json \
"Review the attached code diff following the instructions provided. Output your review as markdown."
Model Selection:
| Model | When to Use |
|---|---|
bailian-coding-plan/qwen3-max-2026-01-23 | Default - strong general review |
bailian-coding-plan/qwen3-coder-plus | Code-focused review |
opencode/gpt-5-nano | Fast lightweight review |
MUST use --format json to get structured output for parsing.
Parse the OpenCode JSON output and extract the review:
VERDICT: (PASS|FAIL|NEEDS_DISCUSSION)Append OpenCode findings to the review summary:
## OpenCode External Review
**Model:** qwen3-max-2026-01-23
**Status:** [PASS | FAIL | NEEDS_DISCUSSION]
### Issues Found
- Critical: [N]
- High: [N]
- Medium: [N]
- Low: [N]
### Details
[OpenCode review output]
| OpenCode Verdict | Action |
|---|---|
| PASS | Record in review summary, proceed |
| FAIL (CRITICAL/HIGH) | Ask user: fix issues or acknowledge and proceed |
| NEEDS_DISCUSSION | Present findings, ask user for decision |
| Timeout/Error | Log error, proceed without OpenCode review (non-blocking) |
OpenCode review is advisory, not a hard gate. Unlike Bee's 6 internal reviewers which are mandatory, OpenCode provides supplementary findings. CRITICAL issues from OpenCode MUST be presented to the user but the user decides whether to act on them.
After review completes:
rm -f /tmp/bee-opencode-review-diff.txt
rm -f /tmp/bee-opencode-review-files.txt
rm -f /tmp/bee-opencode-review-commits.txt
STOP and report if:
| Decision Type | Blocker Condition | Required Action |
|---|---|---|
| OpenCode not installed | CLI binary not found in PATH | Ask user to install or skip |
| No diff available | No code changes between base and HEAD | Skip review (nothing to review) |
| OpenCode execution fails | Non-zero exit code or timeout | Log error, proceed without review |
| Severity | Condition | Required Action |
|---|---|---|
| CRITICAL | OpenCode finds security vulnerability or data loss risk | MUST present to user immediately |
| HIGH | OpenCode finds bugs or logic errors | Present to user, recommend fix |
| MEDIUM | Code quality or architecture concerns | Include in review summary |
| LOW | Style or minor improvements | Include in review summary |
| User Says | Your Response |
|---|---|
| "Skip OpenCode, Bee reviewers passed" | "OpenCode uses a different model - it catches different issues. Running review." |
| "OpenCode is too slow" | "Review typically completes in under 60 seconds. Running review." |
| "I don't trust that model" | "Review is advisory - you decide which findings to act on. Running review." |
| "Just use Bee reviewers" | "OpenCode complements Bee reviewers with a different AI perspective. Running review." |
| Rationalization | Why It's WRONG | Required Action |
|---|---|---|
| "Bee reviewers already covered everything" | Different models catch different issues - shared biases exist | Run OpenCode review |
| "OpenCode model is weaker than Claude" | Model strength is irrelevant - diversity of perspective matters | Run OpenCode review |
| "Results will just duplicate Bee findings" | Unique findings are common across different models | Run OpenCode review |
| "Too much review overhead" | One extra minute for a fresh perspective is minimal overhead | Run OpenCode review |