Skip to main content
OpenSRE provides GitHub-backed architecture audit tools for reviewing repository structure — import boundaries, file placement, size limits, and shim patterns. Use them from the interactive shell (opensre) when you want a structured architecture review before or alongside an incident investigation.

​
Prerequisites

  • GitHub connected via GitHub integration or a token in GITHUB_TOKEN / GH_TOKEN
  • For private repos, ensure the token has repo read scope

​
Typical workflow

  1. Clone the target repository into a fixed temp workspace
  2. Run architecture shell heuristic passes (import, placement, size, shim) via the action agent
  3. Save the full untruncated observation list to disk
  4. Cleanup the clone when finished
In the REPL, describe what you want in plain language:
Or invoke tools explicitly through the action agent.

​
Tools

​
architecture_clone_repo

Returns workspace_root — the path to the clone. Always call architecture_cleanup_repo when finished.

​
architecture_save_observations

Call after cleanup and before the final summarized report. Pass the complete findings from all four shell passes (import, placement, size, shim) — not the short user-facing summary.

​
Verify GitHub access

Expected output includes MCP tool discovery; architecture tools additionally accept GITHUB_TOKEN when GitHub MCP is not configured.

​
Gotchas

  • Always cleanup. architecture_clone_repo leaves a shallow clone on disk until architecture_cleanup_repo runs. The tool description enforces this workflow.
  • Workspace is fixed. Clones go under the architecture workspace in the system temp directory — you cannot point the clone at an arbitrary path.
  • Save the full list. architecture_save_observations expects untruncated pass output. Saving only the short report template loses audit evidence.
  • Session id required. Saving observations needs an active interactive-shell session (~/.opensre/sessions/). Start opensre before running the audit.