codex（Codex 编程助手）

Codex CLI Integration

Two specialized capabilities for different use cases.

Capability 1: Code Generation

Deep code analysis and generation with maximum reasoning power.

When to Use

• Complex code analysis requiring deep understanding
• Large-scale refactoring across multiple files
• Automated code generation with safety controls
• Tasks requiring specialized reasoning models

Default Configuration

• Model: gpt-5.2-codex
• Reasoning: xhigh (maximum thinking depth)

Command Pattern

codex e -m gpt-5.2-codex -c model_reasoning_effort=xhigh \
  --dangerously-bypass-approvals-and-sandbox \
  --skip-git-repo-check \
  -C <workdir> \
  "<task>"

Parameters

• <task> (required): Task description, supports @file references
• -m <model>: Override model (e.g., gpt-5.1-codex, gpt-5)
• -c model_reasoning_effort=<level>: Override reasoning (low/medium/high/xhigh)
• -C <workdir>: Working directory (default: current)

Examples

Basic code analysis:

codex e -m gpt-5.2-codex -c model_reasoning_effort=xhigh \
  --dangerously-bypass-approvals-and-sandbox \
  --skip-git-repo-check \
  "explain @src/main.ts"

Refactoring with custom model:

codex e -m gpt-5.1-codex -c model_reasoning_effort=high \
  --dangerously-bypass-approvals-and-sandbox \
  --skip-git-repo-check \
  "refactor @src/utils for performance"

Multi-file analysis:

codex e -m gpt-5.2-codex -c model_reasoning_effort=xhigh \
  --dangerously-bypass-approvals-and-sandbox \
  --skip-git-repo-check \
  -C /path/to/project \
  "analyze @. and find security issues"

Capability 2: Web Search & Fetch

Online research with web search and page content fetching.

When to Use

• Online research and documentation lookup
• Fetch and summarize specific web pages (GitHub repos, docs, articles)
• Current information retrieval
• API documentation search
• Technology comparison and recommendations

Default Configuration

• Model: gpt-5.1-codex
• Reasoning: high
• Web search: enabled

Command Pattern

codex e -m gpt-5.1-codex -c model_reasoning_effort=high \
  --enable web_search_request \
  --dangerously-bypass-approvals-and-sandbox \
  --skip-git-repo-check \
  "<task>"

Parameters

• <task> (required): Search query or research task
• -m <model>: Override model
• -c model_reasoning_effort=<level>: Override reasoning (low/medium/high/xhigh)
• --enable web_search_request: Enable web search (required for this capability)

Alternative: Config File

Add to ~/.codex/config.toml:

[features]
web_search_request = true

Examples

Fetch GitHub repo:

codex e -m gpt-5.1-codex -c model_reasoning_effort=high \
  --enable web_search_request \
  --dangerously-bypass-approvals-and-sandbox \
  --skip-git-repo-check \
  "Fetch and summarize https://github.com/user/repo"

Documentation search:

codex e -m gpt-5.1-codex -c model_reasoning_effort=high \
  --enable web_search_request \
  --dangerously-bypass-approvals-and-sandbox \
  --skip-git-repo-check \
  "find the latest React 19 hooks documentation"

Technology research:

codex e -m gpt-5.1-codex -c model_reasoning_effort=high \
  --enable web_search_request \
  --dangerously-bypass-approvals-and-sandbox \
  --skip-git-repo-check \
  "compare Vite vs Webpack for React projects in 2024"

Session Resume

Both capabilities support session resumption for multi-turn conversations.

Resume Command

codex e resume <session_id> "<follow-up task>"

Example

# First session (code generation)
codex e -m gpt-5.2-codex -c model_reasoning_effort=xhigh \
  --dangerously-bypass-approvals-and-sandbox \
  --skip-git-repo-check \
  "add comments to @utils.js"
# Output includes: thread_id in JSON output

# Continue the conversation
codex e resume <session_id> "now add type hints"

Notes

• Requires Codex CLI installed and authenticated
• @file syntax references files relative to working directory
• @. references entire working directory
• JSON output available with --json flag for programmatic use
• All commands use --dangerously-bypass-approvals-and-sandbox for automation
• Use --skip-git-repo-check to work in any directory