Read repository instructions and supplied context. Determine assessment-only, runtime QA, or authorized implementation scope. Reuse the user's stated intent.
Map entry points, dependencies, data/state, build/test commands, and ownership. Use rg and focused reads rather than an exhaustive file dump.
State the inferred purpose and material uncertainties. Ask only when an answer changes findings or safety; continue independent static checks meanwhile.
Verify suspected defects with code paths, existing tests, or focused repros. For assessment-only requests, do not automatically install dependencies, launch servers, run browsers, or delegate work. Label untested claims.
When runtime QA is authorized and relevant, consult the test battery. Web UI exploration may use separately installed dogfood. A missing browser does not block static or CLI assessment; report the coverage limit.
Rank findings by impact, evidence, effort, and dependencies. Distinguish bugs, intended behavior, and suggested improvements. Implement only within scope already authorized by the user; otherwise present the proposed work.

Default artifact: app-design-output/recommendations.md, including the app model, findings, tested/untested areas, and next steps. Existing app-model.md and runtime reports remain valid inputs; additional documents are optional.

Question bank: consult only for unresolved intent.

app-design

Existing application assessment