Technical Discovery

Produce a discovery document that maps the technical landscape relevant to a proposed change. The document should give an engineer unfamiliar with the problem enough context to get up and running quickly.

When to Use

• Early in the change process (pitch, PRD, or pre-planning)
• When scoping work for a new feature or change
• When someone needs to understand how existing systems relate to a problem
• Before writing Linear issues or implementation plans

Process

1. Clarify the problem — Ask the user to describe what they're considering. Understand the intent, not just the surface request.

2. Investigate the codebase — Use search tools extensively to find:

   • Relevant models, controllers, services
   • Related features that solve similar problems
   • Data structures and relationships
   • Integration points and dependencies

3. Trace the flow — Use the walkthrough skill if needed to understand how existing systems work together.

4. Document findings — Produce a discovery document (see format below).

5. Identify open questions — Flag things that couldn't be determined or need human input.

Discovery Document Format

# Discovery: [Problem/Proposal Name]

## Context
Brief description of what's being considered and why this investigation was needed.
Keep to 2-3 sentences.

## Relevant Systems
For each relevant system, describe:
- What it does
- How it relates to the problem
- Link to key files

Keep descriptions concise—enough to orient, not to teach.

## Data & Models
Relevant data structures, relationships, and constraints.
Include a Mermaid ERD if relationships are non-trivial.

## Integration Points
Where a solution would need to connect, hook into, or modify existing code.
Be specific—name files, methods, or patterns.

## Constraints & Considerations
Technical limitations, performance concerns, edge cases, or gotchas discovered.
Only include what's relevant to decision-making.

## Open Questions
Things that couldn't be determined or require human input.

Guidelines

Be Comprehensive Yet Concise

• Include enough detail that an unfamiliar engineer can orient quickly
• Exclude tangential information that doesn't inform the problem
• Link to files rather than copying large code blocks
• Use diagrams when they communicate faster than prose

Stay Neutral

• Present findings, not recommendations
• Document what exists and how it relates to the problem
• Leave the "should we?" decision to the reader

Focus on the Problem

• Every section should connect back to the proposed change
• Skip systems that aren't relevant, even if interesting
• If a system is tangentially related, mention it briefly and move on

Surface Unknowns

• Be explicit about what you couldn't determine
• Distinguish between "this doesn't exist" and "I couldn't find it"
• Flag areas that would benefit from human expertise