Agent skill

codex-cli

Operate OpenAI Codex CLI (terminal coding agent) to accomplish software engineering tasks. Use when the user asks to: run codex commands, use codex for coding tasks, execute codex exec for automation, do code review with codex, manage codex sessions (resume/fork), configure codex (config.toml, approval modes, sandbox), use codex cloud, set up MCP servers in codex, or any task involving the `codex` command-line tool. Triggers: codex, codex exec, codex review, codex cloud, codex mcp, codex resume, codex sandbox, openai codex.

Stars 0
Forks 0

Install this agent skill to your Project

npx add-skill https://github.com/zkl2333/skills/tree/main/skills/codex-cli

SKILL.md

Codex CLI

OpenAI's terminal coding agent. Reads, edits, and runs code locally with configurable sandbox and approval controls. Built in Rust, open source.

Quick Reference

```bash

Interactive session

codex "your prompt here"

Non-interactive (scripting/CI)

codex exec "your prompt here" codex exec --json "prompt" # JSONL output codex exec -o result.txt "prompt" # Save last message

Code review

codex review # Review current branch vs main codex review --uncommitted # Review all uncommitted changes codex review --base develop # Review against specific branch codex review --commit abc123 # Review a specific commit

Session management

codex resume # Pick from recent sessions codex resume --last # Continue most recent codex fork --last # Fork most recent into new thread

Cloud tasks (experimental)

codex cloud exec --env ENV_ID "task" codex cloud list codex cloud apply TASK_ID codex cloud diff TASK_ID ```

Key Flags

Flag Short Purpose
`--model` `-m` Select model (e.g. `gpt-5.3-codex`, `o3`)
`--sandbox` `-s` `read-only` / `workspace-write` / `danger-full-access`
`--ask-for-approval` `-a` `untrusted` / `on-failure` / `on-request` / `never`
`--full-auto` Shortcut: `-a on-request -s workspace-write`
`--image` `-i` Attach image(s) to prompt
`--search` Enable live web search
`--cd` `-C` Set the working directory for the `codex` command. This directory will be used as the base for operations by subcommands like `exec`, `review`, `cloud exec`, etc. This flag should be passed directly to the `codex` command. For example: `codex --cd /path/to/project review`.
`--add-dir` Grant additional writable directories
`--config` `-c` Override config value (e.g. `-c model="o3"`)
`--profile` `-p` Load named config profile
`--oss` Use local model provider (LM Studio / Ollama)
`--json` (exec only) JSONL output
`--output-last-message` `-o` (exec only) Write last message to file
`--yolo` Bypass all approvals and sandbox (DANGEROUS)

Approval Modes

  • untrusted (default): Only trusted commands (ls, cat, sed…) run without asking. Others prompt user.
  • on-failure: Run all commands freely. Only prompt if a command fails.
  • on-request: Model decides when to ask.
  • never: Never prompt. Failures go straight back to model.

Sandbox Modes

  • read-only: No file writes, no network.
  • workspace-write: Write only in workspace + specified dirs. Configurable network access.
  • danger-full-access: Unrestricted. Use only in isolated environments.

Slash Commands (Interactive)

Command Purpose
`/model` Switch model mid-session
`/permissions` Adjust approval/sandbox settings
`/review` Run code review
`/diff` Show git diff including untracked files
`/compact` Summarize conversation to free tokens
`/mention` Attach file to conversation
`/new` Start new conversation in same session
`/resume` Resume saved conversation
`/fork` Fork current conversation
`/init` Generate AGENTS.md scaffold
`/mcp` List MCP tools
`/status` Show session config and token usage
`/ps` Show background terminals
`/quit` Exit

Common Patterns

CI/CD Automation

```bash

Run task non-interactively with JSON output

codex exec --json --full-auto "fix all lint errors" | process_results.sh

Structured output with schema

codex exec --output-schema schema.json "analyze this codebase" ```

Code Review in CI

```bash codex review --base main --title "PR #42: Add auth" ```

Multi-Directory Projects

```bash codex --cd ./backend --add-dir ./shared "update the API" ```

Image-Driven Development

```bash codex -i mockup.png "implement this UI exactly" ```

Local OSS Models

```bash codex --oss --CPA ollama -m codellama "refactor this function" ```

Configuration

Config file: `~/.codex/config.toml` (user-level), `.codex/config.toml` (project-level).

For full configuration reference including all keys, MCP server setup, model providers, sandbox permissions, shell environment policies, and OpenTelemetry: see references/config-reference.md.

MCP Server Management

```bash codex mcp list # List configured servers codex mcp add # Add MCP server codex mcp remove # Remove MCP server codex mcp login # Authenticate with server codex mcp logout # Remove server credentials ```

Configure in config.toml:

```toml [mcp_servers.my-server] command = "npx" args = ["-y", "@my/mcp-server"] enabled = true ```

Expand your agent's capabilities with these related and highly-rated skills.

zkl2333/skills

zread

Search and read GitHub repository docs via the Zhipu CodePlan MCP zread server. Use when you need quick repo structure lookup or documentation search without cloning. Triggers: zread, search_doc, repo structure, read_file, GitHub docs search.

0 0
Explore
zkl2333/skills

commit

Generate conventional commit messages following git-cz style by default (emoji between type and subject). Automatically detects project configuration when present. Use when creating git commit messages with Conventional Commits specification. See references/config-detection.md for supported configurations.

0 0
Explore
zkl2333/skills

bark

Send Bark (day.app) push notifications via HTTP. Use when asked to push a short message/alert to iPhone/iPad via Bark/day.app, including scheduled digests and quick “push to phone” actions.

0 0
Explore
zkl2333/skills

caiyun-weather

Get weather data via Caiyun (彩云天气) API through MCP. Use when: user asks about weather, temperature, forecasts, air quality, or weather alerts for any location in China. Supports realtime, hourly (72h), daily (7d), historical (24h), and alerts. Triggers: '天气', 'weather', '气温', '空气质量', 'AQI', '降水', '预报'. NOT for: locations outside China (use weather skill instead).

0 0
Explore
zkl2333/skills

zai-mcp-server

Use Zhipu CodePlan MCP vision server for screenshot/UI understanding: OCR, error screenshot diagnosis, and UI→artifact conversion. Use when you need to analyze images, extract text, or turn UI into code/spec. Triggers: OCR, screenshot, 识图, 提取文字, diagnose error screenshot, ui_to_artifact.

0 0
Explore
zkl2333/skills

amap-maps

高德地图 API(通过 MCP)。地理编码(地址↔坐标)、IP 定位、天气、POI 搜索、路线规划(骑行/步行/驾车/公交)、距离测量。Triggers: '高德', 'amap', '地理编码', 'geocoding', '经纬度', 'POI', '路线规划', '导航'. NOT for: 高德 JSAPI 前端开发(用 AMap-Web/amap-skills).

0 0
Explore

Didn't find tool you were looking for?

Be as detailed as possible for better results