Skip to content

handoff.md

Latest commit

 

History

History
63 lines (49 loc) · 3.33 KB

handoff.md

File metadata and controls

63 lines (49 loc) · 3.33 KB
Raw

Handoff

State: v0.1, accepted by an independent review (prompt 06B, 2026-10-18) that fixed nine defects with regression tests (see Part 1 / "Independent acceptance" in test-report.md). Published 2026-10-19: https://github.com/PatrickNDevelopment/script-audit, branch main; demo https://patrickndevelopment.github.io/script-audit/. CI and Pages are green; the final cross-portfolio audit (2026-10-20) found no P0 or P1 issue. The reviews were separate Claude Code sessions, not human audits.

Run

npm ci
npm run dev          # http://localhost:5177/script-audit/
python -m venv .venv
.venv\Scripts\python -m pip install -r backend\requirements.lock.txt
cd backend; ..\.venv\Scripts\python -m script_audit      # optional backend on 127.0.0.1:8799
npm run verify       # full gate (or scripts\verify.ps1 / scripts/verify.sh)

For the AI path without a real provider, start the backend with SCRIPT_AUDIT_AI_PROVIDER=fake.

What works (covered by tests, see docs/test-report.md)

  • Import (paste, Markdown/text file, workspace JSON, invented sample), editor with revisions, sections/beats outline, word count and duration at an adjustable speed, approval lock.
  • Four separate deterministic audits with rule explanations; preview (side-by-side diff), apply, reject and restore per finding; stale suggestions refused.
  • Claim register with sources and quotes; grounding coverage per sentence.
  • History with exact undo and revert; compare any two revisions; exports (Markdown, narration text, audit report, workspace JSON).
  • Optional AI explanations through the local backend: opt-in per session, strict output contract, unsafe rewrites dropped, usage ledger and rate limit in SQLite with migrations.

What is disabled or absent

  • No AI provider is configured by default. Served from loopback the AI panel checks the backend on its own; on any other host (the static demo) it shows "Not checked" until the user clicks.
  • Hosted demo: https://patrickndevelopment.github.io/script-audit/ (static build). CI and Pages ran green on the first push (2026-10-19).
  • Backlog: SourceTrace import, voiceover timing alignment (docs/backlog.md).

Publication

Published 2026-10-19 with the owner's approval (portfolio decision D-007): public repository https://github.com/PatrickNDevelopment/script-audit (MIT, branch main), GitHub Pages at https://patrickndevelopment.github.io/script-audit/. The first CI run (Linux, Chromium) and Pages deployment were green; a live check loaded the demo, refreshed it and found no console errors and no requests to localhost.

Next action

None required for v0.1; ideas are in docs/backlog.md. The portfolio's final cross-project audit (prompt 13) is next.

Gotchas

  • vite.config.ts sets base: '/script-audit/'; e2e runs the production build on 4178 and two backends on 8799 (fake AI) and 8800 (no AI), with throwaway databases in .e2e-data/.
  • npm run e2e rewrites docs/screenshots; restore with git checkout docs/screenshots unless the UI changed.
  • examples/ is excluded from Prettier on purpose: draft.md contains a double space the demo relies on.
  • The AI rewrite rules exist twice (src/engine/ai.ts and backend/script_audit/ai.py); keep the word lists in step.
  • Changing src/engine/sample.ts changes journey 2 numbers (coverage 1 of 10); unit and e2e tests pin them.