Prompt file imported from pentacon-dashboard/hacker-dashboard (.codex/prompts/generator.md). Copyright stays with the author.

Generator Prompt

Implement the planned change using existing project patterns.

Before editing, read AGENTS.md, .codex/project.md, .codex/instructions/common.md, and the relevant area instruction. For investment analysis, reporting, upload, portfolio, or dashboard work, follow $investment-dashboard.

Implementation Rules

• Prefer focused edits and preserve unrelated dirty worktree changes.
• Keep deterministic metric code separate from LLM narrative.
• LLM output may explain already-computed facts, but must not invent numbers, tickers, causal claims, actions, or personalized investment advice.
• Preserve OpenAPI contracts unless the task intentionally changes API shape.
• If API shape changes, run make openapi and keep shared/openapi.json plus shared/types/api.ts aligned.
• If Router, Analyzer, prompt, gate, or report output contracts change, update golden samples/tests.
• Keep browser MSW scoped to customer/portfolio demo data only; do not add global app mocks for market data, symbol search, quotes, news, watchlist, notifications, settings, uploads, realtime, or Copilot.
• Do not reintroduce NEXT_PUBLIC_COPILOT_MOCK, COPILOT_MOCK, mock_scenario, or route-level mock SSE responses into normal app code.
• Do not modify .claude/, .env*, secrets, local screenshots, Playwright reports, caches, or harness workspaces.

Verification

• Run the narrowest meaningful test first.
• For frontend typed UI/API changes, run cd frontend && npm run typecheck.
• For backend service/API changes, run focused uv run pytest targets.
• For customer-book, upload/import, portfolio API, Docker/Postgres, demo seed, or routed browser-smoke work, run the harness demo preflight and verify / plus linked /clients/<client_id> routes after data loads.
• Report checks not run and why.