Claude Code Subagents
Is a feature of Claude Code
Notes¶
- Used to delegate tasks to
- Avoid messing up with the context window
- Receive system prompt + task from orchestrating agent
- Called by Use a subagent to …
- Built-in agents (documentation)
- General
- Discover
- Plan
Manage¶
- With
/agents, it's possible to create or maintain agents - Defined similarly than Claude Code Skills
Description¶
- One line
- Use Proactively to get the agent called more often
- Use You must tell the agent precisely the filename to pass an argument
Define output¶
- Define an output format, sometimes numbered
- Add Obstacle encountered to avoid future issues
Provide your review in a structured format:
1. Summary: Brief overview of what you reviewed and overall assessment
2. Critical Issues: Any security vulnerabilities, data integrity risks, or logic errors that must be fixed immediately
3. Major Issues: Quality problems, architecture misalignment, or significant performance concerns
4. Minor Issues: Style inconsistencies, documentation gaps, or minor optimizations
5. Recommendations: Suggestions for improvement, refactoring opportunities, or best practices to apply
6. Approval Status: Clear statement of whether the code is ready to merge/deploy or requires changes
7. Obstacles Encountered: Report any obstacles encountered during the review process. This can be: setup issues, workarounds discovered or environment quirks. Report commands that needed a special flag or configuration. Report dependencies or imports that caused problems.
Tool access¶
- Research / read-only subagent -- Only needs Glob, Grep, and Read. Cannot accidentally modify files.
- Code reviewer -- Needs Bash access to run git diff and see what changed, but still doesn't need Edit or Write.
- Styling / code modification agent -- This is where you give Edit and Write access, because the subagent's job is to actually change your code.
Usage¶
- Excel when you need just the answer, not the process
- Research & investigation
- Find a file
- Code review
- Review implemented changes with "fresh eyes"
- Custom system prompts
- (Claude Code ship with rules for code creation)
- Copywriting work
- Style agent (think of loading an external guide file)
To avoid¶
- Expert claims
- Like "you are an expert on Python" → not required as the Claude already have the knowledge, no need of agent
- Multi-step / Sequential pipelines
- Pipelines works when one step is truly independent from the previous ones
- Test runners
- Failing tests won't be as visible than the main thread