← ClaudeAtlas

surveylisted

Audit an entire codebase, or one perimeter of it (front, back, mobile, infra), against the practices current for the stack it actually runs. Reports and prioritizes, fixes nothing, writes nothing. Use when asked for a code audit, a health check, a technical-debt review, or whether the code is over-engineered, duplicated or out of date.
quietmachineworks/qmw · ★ 0 · AI & Automation · score 73
Install: claude install-skill quietmachineworks/qmw
# Survey A survey is what a vessel gets before someone buys or insures it: a surveyor walks the whole boat, hull to rigging, and hands the owner a defect list ordered by what sinks her first. The surveyor repairs nothing. The owner decides what gets fixed, in what order, and what they can live with. **Write nothing, fix nothing.** No files, no config, no commits, no "small obvious cleanups along the way". The output is the report, and every repair is a follow-up the user asks for after reading it. A survey that patched three things on its way through is no longer a survey anyone can trust on a repository they do not own. This is the whole-vessel counterpart to a diff review: not the change of the week, the condition of the codebase as it stands. ## Scope Read the tree's top levels and the manifests, and name the perimeters that actually exist: front, back, mobile, infra, shared libraries, scripts. A perimeter is a body of code with its own stack and its own way of failing, not a folder. Then ask one question: the whole vessel, or which perimeters. Present the perimeters found, with a rough size each, so the user prices the answer. A repository with one perimeter skips the question. A user who already named a scope in their request skips it too. Full survey on a large repository runs one sub-agent per perimeter, each carrying this skill's lenses and returning findings in the report shape below; the main pass merges, dedupes across perimeters, and prioritizes. A finding