- Home
- Skills
- Agents & Orchestration
- dual model airlock
More screenshots
Works with the AI Agents you already use
dual model airlock
Orchestrate cross-model code generation and independent 4-point audits between Gemini, Claude, and Aider CLI agents.
$5
/moSubscription · cancel anytime
Secure checkout via Stripe
- Always the latest version
- Delivered live to your agent via MCP
- Cancel anytime, access ends at period end
dual model airlock
Example session with this skill installed
Use the claude worker to perform a 4-point review on api/auth_handler.py to check for security and concurrency issues.
- Read your context and instructions
- Compiled the dual model airlock
I have initiated the cross-model audit of api/auth_handler.py using the Claude CLI worker. The secondary model has evaluated the code against the 4-point rubric.
Audit Result: FAIL
| Category | Status | Findings |
|---|---|---|
| Correctness & Concurrency | FAIL | Potential race condition in update_session_token at line 42; missing mutex lock during write. |
| Security & Integrity | FAIL | Unsafe string formatting in SQL query at line 108; susceptible to injection if input is not sanitized. |
| Performance & Complexity | PASS | Efficient O(1) lookups; standard library used correctly for cryptographic operations. |
| Verdict | FAIL | High severity defects identified. Remediation required for concurrency and security. |
Execution Receipt
{
"task_id": "v8j2n9k1",
"status": "completed",
"exit_code": 0,
"duration_sec": 14.2,
"receipt_hash": "a1b2c3d4e5f6g7h8",
"worker": "claude"
}
Next steps
- Apply the suggested mutex lock to the
update_session_tokenfunction inapi/auth_handler.py. - Replace the raw string formatting at line 108 with parameterized SQL queries.
- Re-run the review command to verify the fix and flip the verdict to
PASS.
Subscription · always the latest version
- Secure checkout via Stripe
- Cancel anytime
- Delivered live via MCP
About this skill
Single-model coding often suffers from confirmation bias and hallucinated fixes. Dual-Model-Airlock solves this by orchestrating a secondary CLI agent to audit, verify, or execute tasks outside the primary context window.
What it does
- Cross-model delegation dispatches scoped coding tasks to a secondary CLI agent with strict timeout and directory boundaries.
- Independent code auditing runs a 4-point structured rubric through a second model to catch bugs the authoring model missed.
- Consensus verification compares outputs from two independent models to synthesize a single verified patch for complex problems.
- Environment discovery identifies available local CLI agents including Claude, Gemini, Codex, and Aider.
- Integrity tracking generates execution receipts with status codes, duration, and SHA-256 integrity digests for every task.
How it works
- Run a discovery check to identify available CLI coding agents in the local environment.
- Define a bounded task specification including objectives, allowed files, constraints, and verification commands.
- Dispatch the task or file review to a worker agent using the
airlock_runner.pyscript. - Process the returned JSON execution receipt or audit verdict to finalize the code changes.
Frameworks & tools
The skill utilizes a Python-based runner to interface with CLI tools like Claude Code, Google Gemini CLI, Aider, and Codex. It supports local execution and includes an optional REST gateway for multi-threaded orchestration.
Why this beats prompting it yourself
Manual multi-model workflows are prone to context leakage and inconsistent review standards. This skill enforces a deterministic 4-point audit protocol and provides structured JSON receipts, ensuring that code is verified by a fresh model instance without the confirmation bias of the original chat session.
Use cases
- Audit critical security modules using a secondary model before a pull request merge.
- Delegate unit test generation to a background CLI agent while you continue architectural work.
- Resolve complex race conditions by forcing a consensus between two different LLM architectures.
- Automate a ping-pong resolution loop to fix bugs identified during the cross-review process.
Known limitations
Requires Python and local installation of supported CLI agents. Task delegation is restricted to locally accessible files and directory boundaries.
Requires Python and local installation of supported CLI agents like Claude or Aider. Task delegation is restricted to locally accessible files and directory boundaries.
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 any payment.
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 forever
No account required to browse
Trust & safety
Security scan
Security scan passed 2 days ago
Requires
Permissions
Reads the target project file and invokes the user's locally installed CLI coding agent subprocess (claude, gemini, codex, or aider) to run cross-model verification.