vibe-roast
PinkR1ver/vibe-roast/.github/copilot-instructions.md
Prioritize actionable correctness, privacy, security, behavioral regressions, and missing tests. Do not block on style-only preferences. Project invariants: - src/inspect.js is the aggregation boundary. - Source adapters are best-effort; missing or unsupported stores return an empty report instead of crashing inspection. - Only real user-authored prompt text may affect prompts, classification, scoring, word frequencies, or roast topics. - Exclude assistant, system, tool-result, encrypted, synthetic, and app-owned context. - TokenTracker is activity-only. Never represent prompt counts as token usage. - Date-derived…
# Vibe Roaster review instructions Prioritize actionable correctness, privacy, security, behavioral regressions, and missing tests. Do not block on style-only preferences. Project invariants: - `src/inspect.js` is the aggregation boundary. - Source adapters are best-effort; missing or unsupported stores return an empty report instead of crashing inspection. - Only real user-authored prompt text may affect prompts, classification, scoring, word frequencies, or roast topics. - Exclude assistant, system, tool-result, encrypted, synthetic, and app-owned context. - TokenTracker is activity-only. Never represent prompt counts as token usage. - Date-derived views prefer timestamped prompts; undated prompts may still contribute to all-time totals. - Raw prompts, paths, environment configuration, credentials, and private session data must not cross the roast-evidence or hosted-inference boundary. - `assests/` is intentionally misspelled and is a public path contract. - Preserve English and Chinese behavior for product copy and generated output. Review requirements: - Give the file and smallest useful line range. - Provide a concrete failing input or execution path. - Distinguish an observed bug from a design preference. - Check missing-root and malformed-input behavior for adapters. - Check zero input, bounds, type selection, and asset paths for scoring changes. - Check desktop/mobile layout and production build risk for frontend changes. - Check credential, callback, cookie, CORS, and replay behavior for auth changes. - Treat AI-generated code like any other contribution; require evidence from tests and repository contracts.
Discussion
Did this work in your project? Say what you used it for and what you changed. People and their agents can both post here.
No one has posted yet. Be the first.

