mdrubelamin2
UserA discipline layer for Claude Code: terse register, evidence gates on the model's own output, a stop before it builds the wrong thing. Four agents, six skills, one advisory hook — ~2,763 tokens of static context, everything else fires on demand.
Categories
Indexed Skills (7)
codemap
Writes hierarchical codemap.md files across an UNFAMILIAR repo, plus a root atlas and an AGENTS.md section. Expensive, one agent per directory, and it mutates the repo — state the cost and get a yes first.
deep-interview
Interviews the user in rounds of two to four questions before anything is built, then writes a spec file (goal, out of scope, acceptance criteria, verification plan) and STOPS for approval before any code.
deepwork
Runs a migration, rewrite or cross-cutting refactor as staged execution: a written stage map, parallel lanes, one failable check per stage, and a gate between stages.
design
Builds and judges user interfaces against a measured floor, in two modes — replicating a source at fidelity, or originating from a brief — then renders the result and reports what a script measured rather than what it looks like.
review
Reviews a diff across correctness, simplicity, security, tests, schema, API contract and performance at once, behind an evidence gate — every finding quotes file:line with severity and confidence — then fixes what is mechanical.
simplify
DELETES code that should never have been written — speculative abstraction, config nobody sets, hand-rolled standard library — and simplifies the rest with behaviour preserved exactly. Nothing comes out until it is known why it went in.
verification-planning
Plans how to prove a change works or broke nothing: frames the claim, designs an evidence path from the system itself, and requires every check to be able to fail.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.