Feature Request: Per-Agent Directory Access Control for Sub-Agents
Feature Request: Per-Agent Directory Access Control
Currently, sub-agents in Claude Code inherit the project's global .claudeignore file, which limits optimization for specialized workflows.
Use Case Example: A web application with:
- Frontend code in /src/frontend
- Backend API in /src/backend
- Database migrations in /db
- Documentation in /docs
- Test files in /tests
Problem: A frontend-specialist agent loads backend files, database schemas, and test fixtures into its context, wasting valuable context window space on irrelevant code.
Proposed Solution: Allow sub-agents to specify accessible directories:
Option 1 - Custom .claudeignore per agent:
.claude/agents/frontend-agent/.claudeignore
Option 2 - Directory field in agent definition:
name: frontend-specialist
directories: ["/src/frontend", "/src/shared", "/docs/frontend"]
tools: Read, Edit, Grep
Benefits:
- More efficient context usage
- Larger effective working memory per agent
- Better separation of concerns
- Faster agent performance on large codebases
Impact: Enable complex multi-agent workflows on enterprise-scale projects without context exhaustion.
This issue has 4 comments on GitHub. Read the full discussion on GitHub ↗