- Home
- Skills
- Agents & Orchestration
- second opinion
second opinion
Get an automated read-only code review from a different AI model to catch bugs and architectural flaws.
Free
Works with the AI Agents you already use
second opinion
Example session with this skill installed
Review my current changes using codex. I refactored the auth middleware to support JWT rotation.
- Read your context and instructions
- Compiled the second opinion
I have completed a second-opinion review of your JWT rotation changes using Codex. We ran two rounds of review to address token expiration edge cases.
Review Report: JWT Refactor
| Round | Reviewer Findings | Action Taken | Status |
|---|---|---|---|
| 1 | Potential race condition in refresh-token.ts during simultaneous requests. | Added a locking mechanism to the token store. | Fixed |
| 1 | exp claim was not being validated against the clock skew. | Updated validation logic to allow 30s skew. | Fixed |
| 2 | Suggested using a different encryption library for the secret. | Kept current library for consistency with existing infra. | Disagreed |
| 2 | No further issues found in the updated diff. | N/A | Approved |
Final Status: DONE
Reviewer: Codex (OpenAI)
Limits: Usage at 12% of daily quota.
Next steps
- Verify the locking mechanism in the staging environment.
- Merge the changes into the
auth-updatebranch. - Run the full integration test suite to confirm clock skew handling.
Connects securely to your tools. The creator never sees your data.
About this skill
Logic errors and blind spots often slip through when a single agent works in isolation. This skill forces a "second pair of eyes" by sending your diffs to a different AI model for a read-only review. It identifies bugs, architectural flaws, and edge cases your primary agent might have missed, repeating the process until the code is solid.
What it does
- Cross-model review sends uncommitted changes to a secondary provider like OpenAI (Codex), Anthropic (Claude), or DeepSeek.
- Context-aware analysis allows reviewers to open supplementary project files to understand the broader impact of a change.
- Privacy filtering automatically excludes sensitive files like
.env,.pem, and SSH keys from the review diff. - Iterative refinement tracks review rounds and your responses to prevent repetitive feedback and ensure progress.
- Resource monitoring checks API limits before execution to avoid unexpected mid-task failures.
How it works
- Selection identifies the best secondary reviewer based on your primary agent or user preference.
- Analysis runs the review script against uncommitted changes or a specific git base.
- Execution processes the reviewer's feedback, fixing verified bugs while pushing back on incorrect suggestions.
- Conclusion ends the cycle when the reviewer approves (STATUS: DONE) or after a maximum of three rounds.
Frameworks & tools
This skill utilizes a bash-based review script that interfaces with the CLI tools for Codex, Claude, and DeepSeek. It integrates directly with standard Git workflows and requires internet access for API communication.
Why this beats prompting it yourself
Manual cross-checking requires copy-pasting code between windows and losing file context. This skill automates the feedback loop, manages the conversation history between models, and protects your secrets from being sent to external providers.
Use cases
- Reviewing complex refactors before committing to the main branch.
- Getting a high-effort second opinion on security-critical authentication logic.
- Resolving disagreements between the developer and the primary AI agent.
- Sanity-checking large diffs for hidden side effects in legacy codebases.
Known limitations
Requires external CLI tools to be installed and logged in. DeepSeek mode is restricted to diff-only view and cannot read supplemental files. Reviews may take up to 10 minutes for large changes.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.
Requires Codex, Claude, or DeepSeek CLIs installed and authenticated. DeepSeek provider cannot read files outside the diff. Review rounds are capped at three by default to prevent loops.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 18 days ago
- Free with an Agensi account
- Software
Python
Git
Permissions
Allowed Hosts