Menu

Specialist subagents

Specialist subagents are intended for delegated work. Use them when a task has a focused shape and the primary conversation should stay clean.

Codebase agents

Agent Model handler Purpose
codebase/locator scout Finds where files and symbols live. Returns paths only.
codebase/analyzer code_analyzer Explains how code works with precise file references.
codebase/pattern-finder code_analyzer Finds existing patterns and examples to model after.
codebase/coordinator coordinator Coordinates multi-step codebase investigations.

Use these before implementation when you need to understand an unfamiliar area, find the right files, or locate existing conventions.

Code review agents

Agent Model handler Purpose
code-review/coordinator coordinator Orchestrates the review pipeline.
code-review/analyzer code_analyzer Performs review analysis.
code-review/reviewer-security code_analyzer Reviews security issues.
code-review/reviewer-integrity code_analyzer Reviews correctness and robustness.
code-review/reviewer-performance code_analyzer Reviews performance concerns.
code-review/reviewer-consistency code_analyzer Reviews consistency and maintainability.
code-review/scout scout Gathers read-only review context.

Use the coordinator when you want the full review flow. Use an individual reviewer only when you want that specific lens.

Project and documentation agents

Agent Model handler Purpose
project/developer developer Implements scoped project changes and verifies them.
project/document-analyzer small_document_analyzer Answers questions from project docs with evidence.
project/document-writer big_document_analyzer Reserved documentation writer placeholder; currently disabled.

project/developer is best for implementation tasks that are already reasonably scoped. Keep broad architecture questions with a primary agent or an architecture-oriented specialist first.

Research agents

Agent Model handler Purpose
research/web-scout scout Web research and source synthesis.
research/repo-scout scout External repository research through registered reference repositories.

Use these when the answer depends on up-to-date external information or an upstream codebase outside the current worktree.

Permissions and safety

Subagents do not all receive the same tools. Discovery and review agents are intentionally more restricted than implementation agents. This keeps read-only analysis, external research, and write-capable implementation separated by role.

Model routing for all specialist subagents is controlled through Model routing.

On this page
Manual home Built from Markdown

opencode-squad v0.1.0