| name | st-research |
| description | Research phase of the workflow chain. Uses Haiku (simple) or Sonnet (standard) to research the codebase and gather context for a task. Use when the user wants to investigate code before planning. |
Research Phase
Research the codebase and gather all relevant context for the user's task.
Instructions
Agent & Model Selection
Launch the researcher agent (stelite:researcher, read-only tools). Override model based on task complexity:
- Simple tasks (single file, small scope, well-defined) →
model: "haiku"
- Standard tasks (multiple files, moderate scope) →
model: "sonnet"
Use thoroughness level "very thorough".
Parallel Execution
If the task spans multiple independent areas of the codebase, launch multiple Research agents in parallel — one per area. For example, if researching both the API layer and storage layers, launch two agents concurrently. Combine their findings into a single summary.
Research Goals
The research agent(s) should:
- Identify all files relevant to the task
- Understand existing patterns and conventions
- Map dependencies and relationships
- Note any constraints or potential issues
- Gather code snippets and context needed for planning
After the agent(s) complete, present a structured summary of findings:
- Relevant files — paths and their roles
- Patterns — conventions and approaches used in the codebase
- Dependencies — what connects to what
- Constraints — things to watch out for
- Key context — important code snippets or decisions
