A customized Claude Code setup with domain-specific skills tuned to how we actually build, ship, and debug at Progcap.
See the skills Sign in with GoogleGeneric Claude is smart. Mitroo is smart and opinionated about your stack.
Pre-loaded with Progcap conventions, repos, and Metabase schemas — no re-explaining the basics each session.
Real shipping motions: TDD, QA loops, ship + deploy, visual review — all one command away.
Each skill replaces a 10-message prompt with one verb. You spend time reviewing, not steering.
Eight commands that generic Claude does not have. Each one collapses a multi-step workflow into a single action.
Query TAT, referbacks, NPA, drop-offs — with our schemas already understood. Generic Claude would need a 50-line schema dump first.
Drives a real browser, finds bugs, fixes them in source, re-verifies. Generic Claude can only suggest fixes.
Tests, version bump, changelog, commit, push, PR — in one go. Generic Claude stops at "git commit -m".
Four-phase forensic loop with the Iron Law: no fix without root cause. Generic Claude jumps to patches.
Live screenshots, spacing audits, AI-slop detection, then iterative fixes. Generic Claude cannot see the page.
Adversarial diff review by a different model. Generic Claude only marks its own homework.
Type checks, lint, tests, dead code — weighted into one score with trends. Generic Claude has no memory between runs.
Drives the docs and onboarding flow, times TTHW, scores friction. Generic Claude only reads README text.
Same model underneath. Wildly different leverage on real product work.
| Capability | Generic Claude | Mitroo |
|---|---|---|
| Onboarding context | Re-explain every session | Pre-loaded conventions |
| Domain queries (TAT, NPA, referbacks) | ✗ Manual SQL | ✓ /metabase |
| QA loops | Suggests fixes only | Test · fix · verify in one run |
| Ship / deploy flow | Stops at commit | PR + deploy + canary |
| Visual review | ✗ No eyes | ✓ Screenshots + fixes |
| Second opinion | Marks own work | /codex adversarial review |