Skip to content

Product Definition

Product Thesis

AgentTrace is the local audit layer for AI-assisted coding. It records the task, repository state, code changes, tests, reviews, and final evidence behind an AI coding run.

User Persona

The primary user is an engineering lead or senior developer adopting coding agents in a professional software team. They want the productivity benefits of AI assistance without losing accountability, review discipline, or operational visibility.

Enterprise Problem

AI coding tools can make meaningful changes quickly, but teams often lack durable answers to basic governance questions:

  • Which tool was used?
  • What task did it attempt?
  • What files changed?
  • Were tests or builds run?
  • Was the diff reviewed?
  • What evidence supports the final claim?

AgentTrace gives teams a lightweight local record before they invest in centralized dashboards or enterprise integrations.

MVP Scope

The MVP is a CLI that creates a structured run folder under .agenttrace/runs/<run-id>/ and captures:

  • task description
  • branch and commit metadata
  • git status
  • git diff
  • test/build commands and outputs
  • review notes
  • a Markdown report

Non-Goals

  • No hosted service.
  • No remote repository creation or pushing.
  • No automatic source control commits for user code.
  • No policy enforcement engine.
  • No dashboard in the MVP.
  • No claims about whether AI-generated code is correct beyond recorded evidence.

Future Expansion Paths

  • GitHub PR comment/report integration.
  • MCPGuard integration for tool-use policy and safety evidence.
  • EvalOps integration for benchmark and regression tracking.
  • PromptOps contract validation.
  • Risk scoring for sensitive files and failing evidence.
  • Team dashboard for aggregated run history.