AgentSkillsCN

codex

将编码任务委托给Codex CLI（通过JetBrains AI的OpenAI gpt-5-codex）。当任务涉及可外包的重复性代码生成、重构或分析时使用。

SKILL.md
--- frontmatter
name: codex
description: Delegate coding tasks to Codex CLI (OpenAI gpt-5-codex via JetBrains AI). Use when a task involves repetitive code generation, refactoring, or analysis that can be offloaded.

Codex Skill - JetBrains AI Integration

Use Codex CLI to delegate coding tasks that benefit from parallel AI processing.

When to Use Codex

GOOD for Codex:

  • •Large-scale refactoring across many files
  • •Repetitive code generation (boilerplate, tests)
  • •Code analysis and documentation
  • •Pattern-based transformations
  • •Exploring unfamiliar codebases

KEEP in Claude:

  • •Complex architectural decisions
  • •Security-sensitive code
  • •Tasks requiring deep context understanding
  • •Interactive problem-solving with user

Configuration

Codex is pre-configured with JetBrains AI staging:

  • •Provider: jbai-staging
  • •Models: gpt-4o-2024-11-20, o1-2024-12-17, o3-mini-2025-01-31

Usage Patterns

Quick Analysis (read-only)

bash
codex exec -c model_provider=jbai-staging --model "gpt-4o-2024-11-20" \
  --sandbox read-only \
  "Analyze the codebase and list all REST endpoints" 2>/dev/null

Code Generation (workspace-write)

bash
codex exec -c model_provider=jbai-staging --model "gpt-4o-2024-11-20" \
  --sandbox workspace-write --full-auto \
  "Generate unit tests for UserService.kt" 2>/dev/null

Heavy Reasoning (o1/o3 models)

bash
codex exec -c model_provider=jbai-staging --model "o3-mini-2025-01-31" \
  --sandbox read-only \
  "Review security vulnerabilities in auth module" 2>/dev/null

Resume Previous Session

bash
echo "continue with the refactoring" | codex exec --skip-git-repo-check resume --last 2>/dev/null

Decision Framework

Before delegating to Codex, consider:

FactorUse CodexKeep in Claude
File countMany filesFew files
Task typeRepetitive/mechanicalCreative/strategic
Context neededSelf-containedRequires conversation context
User interactionNone neededBack-and-forth required

Execution Protocol

  1. •Announce intent: Tell user you're delegating to Codex and why
  2. •Choose model:
    • •gpt-4o-2024-11-20 for fast tasks
    • •o1-2024-12-17 / o3-mini-2025-01-31 for complex reasoning
  3. •Choose sandbox:
    • •read-only for analysis
    • •workspace-write for edits
  4. •Run command with 2>/dev/null to suppress thinking tokens
  5. •Report results back to user
  6. •Offer resume: "Say 'codex resume' to continue this session"

Safety Rules

  • •Always use --skip-git-repo-check flag
  • •Default to read-only sandbox unless edits needed
  • •Never use danger-full-access without explicit user permission
  • •Suppress stderr with 2>/dev/null by default