Subagent Parallel Dispatch

In large codebases, a single linear conversation can quickly suffer from context bloat or conflicting goals. JeikCode provides a native subagent parallel dispatch runtime, allowing the primary agent to break down research, refactoring, and multi-file tasks into isolated subtasks.

---

Subagent Architecture & Roles

JeikCode defines two distinct subagent types with strict permission boundaries:

1. Read-Only Research Subagents (explore)

• Permissions: Read-only sandbox environment; cannot write or modify any files on disk;
• Use Cases: Tracing cross-file call chains, analyzing dependencies, evaluating alternative architectures;
• Advantage: Extremely fast and safe. Multiple explore subagents can run concurrently without side effects, returning summarized insights directly to the main agent.

2. Write-Scoped Execution Subagents (worker)

• Permissions: Confined to an explicit file path whitelist (scope);
• Use Cases: Implementing a dedicated feature or module (e.g. updating schemas or refactoring a service layer);
• Scope Sandboxing: Every worker subagent must specify its allowed relative path globs (e.g. ["src/auth/**", "Cargo.toml"]). Any attempt to edit or write files outside this scope is blocked at the runtime level.

---

Parallel Workflow Diagram

               ┌───────────────────────┐
               │ User Task / Prompt    │
               └──────────┬────────────┘
                          │
                          ▼
               ┌───────────────────────┐
               │      Main Agent       │
               └──────────┬────────────┘
                          │ Parallel Dispatch
             ┌────────────┴────────────┐
             ▼                         ▼
   ┌───────────────────┐     ┌───────────────────┐
   │ Subagent [explore]│     │ Subagent [worker] │
   │ Research Module A │     │ Implement API B   │
   │ (Strict read-only)│     │ (Scoped to B)     │
   └─────────┬─────────┘     └─────────┬─────────┘
             │ Summarized findings     │ Changes completed
             └────────────┬────────────┘
                          ▼
               ┌───────────────────────┐
               │ Aggregate & Deliver   │
               └───────────────────────┘

---

Key Benefits

1. Context Cleanliness: Subagents operate in their own temporary contexts. Heavy command outputs and intermediate file reads never pollute the primary agent's token window.
2. Speed & Concurrency: Independent research tasks execute in parallel, substantially shortening total turnaround time.
3. Safety Isolation: Enforcing explicit write scopes on worker subagents eliminates unintended accidental edits across unrelated code files.