Runs OpenAI Codex CLI non-interactively to obtain second-opinion code reviews and design feedback.
npx skills add https://github.com/MaxWolf-01/agents --skill codex-maxwolf-01
Please help me install this Agent Skill. Skill: codex Source: https://github.com/MaxWolf-01/agents/tree/main/mx/skills/codex Command: npx skills add https://github.com/MaxWolf-01/agents --skill codex-maxwolf-01
codex exec in read-only sandbox mode with full disk read access, so Codex can inspect the codebase without modifying anything. - Background job management: Launches Codex as a named background job with a deadline, then reads the final answer from an output file. - Prompting guidance: Provides concrete instructions for orienting a zero-context model—pointing it at CLAUDE.md, decisions records, and ticket files, and requesting structured output. - Use Case: While debugging a flaky test, ask Codex to review the relevant files and list the top five likely causes with fixes, then compare its answer against your own analysis. ## Quick Start Ask the agent to get a second opinion from Codex on the current design or bug, and it will run Codex in the background and report back the findings.High-intent search queries and answers about installing and using this skill.
Run OpenAI's Codex CLI non-interactively with codex exec in read-only sandbox mode, passing a prompt that names the files to review and the question to answer. The final response is written to an output file you read after the job completes.
Launch it as a named background job with a deadline, for example job run codex-<slug> with codex exec and an output file flag, then wait on the job and read the result file. The deadline prevents a wedged run from blocking the session.
Yes. Use the read-only sandbox flag together with disk-full-read-access permission so Codex can read any file on disk but cannot write. This guarantees no conflicts with your working tree during review.
Codex starts with zero session context, so vague prompts produce vague answers. Point it to project documentation, name the exact files to review, state your constraints, and request structured output such as a numbered list of issues with fixes.
Raise the model reasoning effort setting only for especially reasoning-intensive tasks such as a subtle bug or a difficult design question. For routine reviews, the default configuration is sufficient and avoids unnecessary latency.
DharambirAgrawalBundle prompts with selected files into context-aware requests for code review.
pcortesClassify peer feedback on engineering specs into ACCEPT, REJECT, DEFER, or PARTIAL decisions.
shakhovskiya-createRun Ruff, MyPy, Bandit, Pip-audit, Radon, Vulture, and Mutmut audits on Python projects.
mscully4Review work products with a different AI model for a second opinion.