docs: Sync docs and agent workflows to completed stack

[mlarsen-source]

Summary

Root README, backend README, and backend/.env.example were updated to reflect the fully-implemented pipeline: LLM feedback correctly attributed to Groq (openai/gpt-oss-20b), both /feedback/generate and /feedback/generate-session marked done with full request/response documentation, scorecard components and lib helpers added to the repo structure, deploy rows changed to TBD, and the .env.example header corrected to stop stating all three keys are required. All eight agent workflow documents were revised to remove patterns inapplicable to this project — Next.js Route Handlers, Zod, E2E tests, pnpm test runner, flag registry, and DB query references — and to add Interview Coach–specific guidance covering FastAPI endpoint patterns, Storybook as the primary coverage mechanism, both feedback modes in QA checklists, and project-accurate file paths throughout.

What was corrected

Stale content type	Files fixed
LLM provider, endpoint status, deploy plans, repo structure	README.md, backend/README.md
Env var requirement header contradicting per-key comments	backend/.env.example
Non-existent patterns (Route Handlers, Zod, E2E, pnpm test, flag registry, DB queries)	8 agent workflow docs
Missing project-specific guidance (FastAPI patterns, feedback modes, Storybook coverage)	8 agent workflow docs

File-by-file changes

Root documentation

• README.md: Removed "(Coming)" from steps 8–9, corrected LLM provider from "Anthropic Claude (coming)" to "Groq (openai/gpt-oss-20b)", added Node.js 18+ prerequisite, expanded repo structure with all implemented scorecard components and lib helpers, changed deploy rows to TBD, and added a full backend endpoints table to reflect the completed pipeline state.

Backend documentation

• backend/.env.example: Changed line 3 header from "All three keys are required for the full pipeline to function" to accurately state that only GROQ_API_KEY is required and the remaining keys are reserved, eliminating a contradiction with the per-key comments below it.
• backend/README.md: Marked /feedback/generate as done, added /feedback/generate-session as done, updated structure section with both LLM endpoints and schemas.py, and added full request/response documentation blocks for both feedback endpoints matching the Pydantic schemas in backend/services/llm/schemas.py.

Agent workflows

• docs/agent-workflows/branch-change-impact-audit.md: Renamed "Feature-flag involvement" to "Env-var or behavioral mode changes" and scoped the trigger to NEXT_PUBLIC_* behavioral gating only, because featureFlag/featureGate patterns do not exist in this project.
• docs/agent-workflows/code-quality-review.md: Removed the await in API route handlers bullet and Zod validation check (no Route Handlers exist), replaced with a FastAPI-specific client-side fetch error handling check; removed Priority 10 (new API route security) entirely; updated Priority 4 to reference Pydantic/types.ts instead of Zod; scoped pnpm test instruction to "not configured — skipped."
• docs/agent-workflows/feature-flag-gating-review.md: Added a "Project context (Interview Coach)" section stating no formal flag system exists, defining when to apply the workflow (NEXT_PUBLIC_* behavioral gating or complex mode branches), and noting the flag registry check does not apply.
• docs/agent-workflows/feature-implementation-planning.md: Updated plan item 2 to name external services (Groq Whisper, Orpheus, LLM, wav2vec2); replaced generic file manifest groupings with actual project paths (backend/services/<service>/router.py, schemas.py, lib/interview-coach/, lib/speech/, etc.); added feedback mode verification to item 6.
• docs/agent-workflows/figma-design-to-code.md: Added a note that the storybook script is not yet in frontend/package.json but stories are required by convention regardless, to prevent agents from skipping story creation because the script is absent.
• docs/agent-workflows/manual-qa-checklist-generator.md: Replaced the generic "feature flags, environments, or auth modes" language in section 4 with explicit perQuestion/endOfInterview mode checks and full scorecard component path verification, so generated QA checklists always cover both feedback pipeline paths.
• docs/agent-workflows/pre-merge-full-review.md: Updated Phase 4 deterministic floor from pnpm test to Storybook story coverage check; updated Phase 5 trigger from featureFlag/featureGate patterns to NEXT_PUBLIC_* behavioral gating; replaced pnpm test --findRelatedTests in the edit protocol with pnpm typecheck and pnpm lint; replaced the DB query invariant example with a FastAPI response shape example.
• docs/agent-workflows/test-suite-quality-review.md: Hardened the "no test runner" note to a definitive statement at the top, removed the E2E test review section, renamed "Breakdown by test type" to "Breakdown by coverage type," and established Storybook stories as the primary coverage mechanism to stop agents from searching for test infrastructure that does not exist.