- Home
- Skills
- Agents & Orchestration
- codex subagent
codex subagent
Delegate independent code reviews and research tasks to a local Codex subagent with persistent threading.
$1
Works with the AI Agents you already use
codex subagent
Example session with this skill installed
Have Codex review the changes in ./src/auth. Hunt for security bugs and race conditions in the login flow. Start a new thread.
- Read your context and instructions
- Compiled the codex subagent
I have delegated the security review of the ./src/auth directory to a local Codex subagent. A new persistent thread has been initialized.
Codex Security Review Findings
| Severity | Finding | Location | Description |
|---|---|---|---|
| High | Potential Timing Attack | auth.ts:42 | The comparePassword function returns early on mismatch, allowing for timing-based enumeration. |
| Medium | Missing Rate Limit | login.ts:12 | The login endpoint does not implement exponential backoff or basic rate limiting for failed attempts. |
| Low | Verbose Error | auth.ts:88 | Error response indicates whether the user exists or the password was wrong, leaking account data. |
Subagent Details
- Thread ID:
ctx_9821-ab92 - Sandbox:
read-only - Status: Analysis complete, awaiting follow-up instructions.
Next steps
- Run
followupfor threadctx_9821-ab92to request a constant-time comparison implementation for the timing attack. - Verify the rate-limiting finding by checking if
middleware/rate-limiter.tsis correctly applied to the login route. - Update the login controller to return a generic "Invalid credentials" message.
Connects securely to your tools. The creator never sees your data.
About this skill
Development often suffers from confirmation bias, especially during self-review or complex debugging. The Codex Subagent skill enables multi-agent delegation by bridging your environment to a local OpenAI Codex CLI. It allows you to spin up independent subagents that can audit your work, perform adversarial reviews, or execute isolated research tasks without cluttering your main conversation context.
What it does
- Persistent threading allows for multi-turn conversations where the subagent remembers previous context and instructions.
- Adversarial review triggers an independent audit of your code to find edge cases, contract violations, and logic bugs.
- Granular sandboxing controls access levels, from read-only investigations to workspace-write permissions for automated fixes.
- External verification provides a second opinion on complex implementations to reduce the risk of biased errors.
How it works
- Initiate a new session using the
asktool, providing a self-contained prompt and specific working directory. - Review the output and the generated
threadIdto determine if further investigation is required. - Iterate using the
followuptool to challenge findings, request evidence, or refine the subagent's output within the same thread. - Relay verified results back to the main workflow, citing the subagent for transparency.
Frameworks & tools
This skill requires the dsh-codex-bridge MCP server, OpenAI Codex CLI, and Node.js 18+. It is compatible with MCP-enabled environments like Claude Code, Cursor, and DSH.
Why this beats prompting it yourself
Manual prompting within a single session leads to context bloat and model bias. This skill offloads specific cognitive tasks to a fresh instance with its own persistent state, ensuring a truly independent critique of your code.
Use cases
- Run an adversarial code review on a git diff before submitting a PR.
- Delegate a background debugging task while you continue feature development.
- Cross-check a complex architectural decision against a second model's logic.
- Execute sandboxed file operations or test runs via a separate agent thread.
Known limitations
Codex cannot see the primary conversation context unless explicitly included in the prompt. Requires local installation of the Codex CLI and the dsh-codex-bridge.
Codex lacks access to the main chat history unless passed manually. Requires local Codex CLI and Node.js 18+. Sandbox levels must be configured in the MCP settings.
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
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 1 month ago
Requires