Discover Codebase Enhancements

Overview

Spend significant time crawling and analyzing the codebase to surface high-impact improvements. Center findings on the jobs-to-be-done of the codebase, developers, end users, and AI agents working in the repo.

Inputs (ask if missing, max 5)

• Target area or scope (whole repo or specific modules)

• Primary user jobs-to-be-done and business goals

• Known pain points or incidents

• Constraints (time, risk tolerance, release window)

• Evidence sources allowed (tests, metrics, logs)

Jobs-to-Be-Done Lens

• Codebase: reliability, simplicity, maintainability

• Developers: speed, clarity, safe changes

• End users: correctness, performance, usability

• AI agents: discoverability, consistency, explicit patterns

Workflow

• Deep crawl

• Read architecture docs, READMEs, key modules, and tests.

• Search for hotspots (TODO/FIXME, large files, duplication, complex flows).

• Evidence gathering

• Note error-prone areas, missing tests, performance risks, and coupling.

• Capture references to files/functions and concrete symptoms.

• Opportunity synthesis

• Group findings by theme: correctness, performance, DX, architecture, tests, tooling.

• Impact scoring

• Rate impact, effort, risk, and evidence strength.

• Ranked recommendations

• Present top enhancements with rationale and expected outcomes.

Output Format

Codebase Enhancement Discovery

Context Summary

[1-3 sentences]

JTBD Summary

• Codebase: ...
• Developers: ...
• End users: ...
• AI agents: ...

Evidence Sources

• Files/modules reviewed: ...
• Patterns searched: ...
• Tests or metrics considered: ...

Ranked Enhancements

1. [Enhancement]
   • Category: ...
   • Impact: high | Effort: medium | Risk: low | Evidence: moderate
   • Rationale: ...
   • Affected areas: ...

Quick Wins

• ...

Open Questions

• ...

Quick Reference

• Spend more time exploring than feels necessary.

• Prefer evidence-backed findings over speculation.

• Center recommendations on user and developer outcomes.

Common Mistakes

• Skimming without enough code context

• Listing fixes without evidence or impact scoring

• Ignoring AI agent or developer workflows

• Recommending changes that fight existing architecture