AGENT SKILL

codex-agent

Run autonomous coding tasks with OpenAI Codex CLI — features, bug fixes, refactors, and PRs.

License 许可
MIT
所属仓库
Folken2/nuvel
最近更新
2026-11-05
Stars
★ 2
Forks
1

01 前言 Front Matter

name
codex-agent
description
Run autonomous coding tasks with OpenAI Codex CLI — features, bug fixes, refactors, and PRs.

03 SKILL.md 正文

Codex CLI Agent Workflows

Trigger Conditions

  • User requests: "build feature X", "fix bug Y", "refactor Z"
  • Need to implement a feature from a spec or issue
  • Generating pull requests from task descriptions
  • Running Codex in agent mode for autonomous code changes

Prerequisites

  • Codex CLI installed and authenticated (codex-setup skill)
  • Nuvel OrgMemory MCP server configured
  • A git repository with clean working tree
  • Clear task description or issue reference

Steps

1. Prepare the Task Context

Gather relevant context before running the agent:

# Check OrgMemory for related decisions and architecture
codex exec "Search OrgMemory for any design decisions or docs related to <feature/component>"

Create a task spec file (or use an existing GitHub issue):

cat > .codex/task.md << 'EOF'
# Task: <Brief Title>

## Description
<Detailed description of what needs to be built or changed>

## Acceptance Criteria
- [ ] Criterion 1
- [ ] Criterion 2

## Related
- OrgMemory: <link to relevant knowledge graph entry>
- Issue: #<number>
EOF

2. Run Codex in Agent Mode

Basic feature implementation:

codex exec --approval-policy on-request \
  "Read .codex/task.md and implement the described feature. \
   Check nuvel-orgmemory for any relevant architecture decisions first. \
   Write tests, implement the changes, and ensure all existing tests pass."

Bug fix workflow:

codex exec --approval-policy on-request \
  "Fix the bug described in .codex/task.md. \
   First reproduce the issue, then identify the root cause, \
   implement the fix with tests, and verify the fix resolves the issue."

3. Review and Iterate

After Codex completes:

# Review the diff
git diff

# Run the test suite
npm test  # or your project's test command

# If issues found, ask Codex to fix them
codex exec "The tests are failing with: <error>. Fix the implementation."

4. Commit and Create PR

git add -A
git commit -m "feat: <description>

Implemented via Codex CLI with Nuvel OrgMemory context.
Closes #<issue-number>"

# Push and create PR
git push origin HEAD
gh pr create --title "feat: <description>" --body "Automated implementation via Codex CLI."

5. Update OrgMemory

After merging, persist new knowledge:

codex exec "Update nuvel-orgmemory with the new design decisions \
  and patterns introduced in this implementation. \
  Reference the merged PR for context."

Advanced: Multi-File Refactors

For large refactors spanning many files:

codex exec --approval-policy on-request \
  "Perform a refactor of <component> to <new pattern>. \
   Step 1: Analyze all files that need changes. \
   Step 2: Run the existing test suite to establish baseline. \
   Step 3: Apply changes file by file, running tests after each. \
   Step 4: Update any documentation."

Pitfalls

  • Large changes: For changes spanning 20+ files, break into multiple Codex sessions to avoid context overflow.
  • Approval policy: Always use on-request (not auto) for production codebases. Auto-mode should only be used in sandbox environments.
  • MCP context: If Codex is not using OrgMemory, verify the MCP server is running and reachable.
  • Git state: Always ensure a clean or committed working tree before running agent mode. Codex may amend unexpected files.

Verification

  1. All tests pass: npm test
  2. Linter clean: npm run lint
  3. Diff is scoped to the intended changes: git diff main...HEAD
  4. OrgMemory updated with new knowledge

04 同仓库更多 Skills