The engineering reviewer that remembers — every finding traced to a real rule, tracked from open to resolved, judged against the repo you actually wrote.

Debuggatha reads your repository before it opens its mouth — stack, dependencies, linter/formatter config, ADRs, README intent — and reviews your diff, a file, or the whole workspace against what it finds. Every finding cites why it exists: a rule in a Review Pack, a convention your own repo declares, a recognized standard (OWASP, WCAG), or evidence in the code itself. No source, no finding.

Why this isn't another AI chat panel

Copilot, Cursor, and Claude Code are built to generate code — fast, capable, and that's their job. Debuggatha has exactly one job: review, after something else (you, or one of those tools) has written the code. It doesn't answer from general training-data intuition — it builds a RepositoryContext from your actual stack and rules first, then reviews against that. And it remembers: findings persist across sessions in a versioned ledger with a real lifecycle (open → acknowledged → resolved → dismissed → reopened), so a review isn't a wall of text you re-read from scratch tomorrow.

What you get in VS Code

Findings Ledger view — every open, resolved, and dismissed finding, filterable, with one-click resolve/dismiss/reopen.
Repository Intelligence view — see what Debuggatha detected about your stack before you ask for a review.
Review commands — Debuggatha: Review Entire Workspace, Review Current File, Review Selection, Review Pending Changes (diff-scoped), all from the Command Palette or the editor toolbar.
Configurable review depth and packs — quick, full, or architectural, with default Review Packs tuned per project.

Getting started

Install the extension.
Open a workspace with a repository (git-initialized).
Run Debuggatha: Review Entire Workspace once to seed the Findings Ledger, or Review Pending Changes to review just your current diff.
Findings show up in the Debuggatha activity bar view — click one to jump to its location.

Debuggatha runs its default review through your MCP host's own configured model (no keys to manage) or a local runtime (Ollama, LM Studio) if you set one — see debuggatha.runtime in Settings.

Also available as

MCP server (@debuggatha/mcp) — for Claude Code, Claude Desktop, or any MCP-speaking host.
CLI (@debuggatha/cli) — for terminal use and CI, no editor required.

See the project README for the full picture, or CONTRIBUTING.md to work on Debuggatha itself.