Add bad-output feedback loop to Prompt Token Saver

[alekslinde]

Summary

• Users can now report bad compression results directly from the tool — a "Bad output? Report it" link appears below the output after each compression, expanding to an optional 280-char note field before submitting
• Reports are stored in a new FeedbackStore Durable Object (SQLite) so they can be retrieved as a training corpus to improve the compression dictionaries over time
• Abuse prevention is layered: origin check (lindetoolbox.com only), payload size limits (10–2000 chars), per-IP+content-hash deduplication per day (same prompt can't be submitted twice by the same person), and a hard cap of 10 reports/IP/day
• No PII is retained — IPs are SHA-256 hashed before storage and the hash is not stored alongside raw text
• Admin corpus retrieval via GET /feedback requires a Bearer token (FEEDBACK_SECRET Worker secret, already set)

Test plan

• Compress a prompt → "Bad output? Report it" link appears below output
• Click link → form expands, button hides
• Submit with no note → 200, success message shown, form collapses
• Submit the same prompt again → 409 "Already reported." shown
• Submit 10 reports from same IP → 11th returns 429 "Too many reports today."
• curl https://lindetoolbox.com/feedback with no token → 401
• curl https://lindetoolbox.com/feedback -H "Authorization: Bearer <secret>" → JSON rows array
• Click Clear → feedback area hides; re-compressing shows fresh form
• Request from non-lindetoolbox origin → 403

[cloudflare-workers-and-pages]

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status	Name	Latest Commit	Updated (UTC)
✅ Deployment successful! View logs	lindetoolbox	598e662	Jun 11 2026, 01:48 PM