Skip to content

Feedback - Research Agent: Limited Analysis Depth and Inconsistent Output File Organization #10

Description

@bakes82

Description

The research agent is currently performing only high-level analysis and lacks the depth needed for comprehensive codebase evaluation. Additionally, output files are being scattered across different directories in an inconsistent manner.

Desired Functionality

Deeper Analysis Capabilities

The agent should perform more granular, service-level analysis similar to the following manual workflow:

Example Use Case:

  • Codebase contains multiple solutions (Web + API)
  • API layer contains ~30 individual services
  • Manual approach uses Haiku for quick service-level extraction across projects
  • Builds a mapping YAML file tracking:
    • Dependencies between services
    • Project references
    • Performance analysis flags
    • Security analysis flags

Suggested Analysis Features

  1. Cloud Migration Planning

    • Generate upgrade paths to Azure/AWS/Google Cloud
    • Include HA/DR considerations using containerization
    • Provide good/better/best approaches
  2. Microservices Transformation

    • Analyze what it would take to convert services to microservices (e.g., AWS-based)
    • Offer tiered recommendations (good/better/best approaches)

File Organization Issues

Output files are currently being placed inconsistently across multiple locations:

Current Behavior

/test/architecture-diagrams.md
/test/.orchestr8/docs/development/architecture-review-report<date>.md
/codeRepos/technical-debt.md
/codeRepos/tech_debt.md
/codeRepos/security_arch.md
/codeRepos/arch_map.md
/codeRepos/.orchestr8/<rest of files>

Expected Behavior

All output files should be placed in the Claude Code calling directory with session-based organization:

/test/.orchestr8/session_<timestamp>/
  ├── architecture-diagrams.md
  ├── architecture-review-report<date>.md
  ├── technical-debt.md
  ├── tech_debt/
  ├── security_arch/
  └── arch_map/

Rationale: Analysis is initiated from /test on projects located in /codeRepos. Documentation should not be written into the code repositories themselves, but rather kept in the calling directory with proper session isolation.

Environment

  • Running Claude Code from: /test
  • Analyzing projects in: /codeRepos
  • Multiple analysis runs on different projects from the same calling location

Context Note

I work in Architecture, and we are doing analysis on projects for performance, security and other optimizations with also DR/HA strategy and modernization plans to move legacy applications to the cloud. So, we are looking to do some deeper level analysis of these older applications.

Metadata

Metadata

Assignees

Labels

enhancementNew feature or request

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions