Changelog
What's new in fdbck, newest first.
v1.0.0
Close the loop — everywhere
July 2026- Notification channels: pipe new feedback, opened PRs, merged fixes, and failures to Slack, Discord, Telegram, or any webhook (HMAC-signed, so Zapier/n8n/your own service work too). Configure up to 10 in Account → Notifications, each with per-event toggles and a Send-test button.
- Reporters can now leave an email in the widget — and get exactly one “Your feedback shipped” email when the fix merges, with a one-click opt-out. Never shown publicly, never sent to the AI agent.
- fdbck speaks MCP: connect your inbox to Claude Code or Cursor and let your coding agent list, inspect, approve, and retry feedback as tools — authenticated with your existing worker token. See the new MCP docs.
- The agent learns from you: when you merge one of its PRs with edits, fdbck captures what you changed and feeds it into future runs on that project. A new confidence-calibration strip in the Track Record card shows whether high-confidence items really merge clean.
- Live run checkpoints: an in-flight fix now shows exactly what the agent is doing (“editing src/billing.tsx · 12 reads · 3 edits”) instead of a spinner — in the dashboard and via MCP.
- Pricing is public: flat, cheap, and unlimited where it counts — AI triage and fixes are never metered, on any tier. Free covers your first project; Pro is a flat $19/mo ($15/mo yearly).
v0.9.0
A premium feedback experience for your users
July 2026- Configure what users see after they submit — a custom confirmation message, footer links, an optional redirect, and a “Powered by fdbck” toggle — all from Settings → Widget messaging, no code changes.
- Every submitter gets a private “Track your feedback” status link: a clean, honest timeline from Received to Shipped, with no account needed.
- New public feedback board: publish the items you choose to a curated, embeddable roadmap (Planned / In progress / Shipped) with anonymous upvotes. Nothing is public until you publish it.
- Tougher on PII and abuse: smarter redaction (real card numbers, IPs, and emails go; your order ids and version numbers stay), a live character counter, and untrusted feedback is always treated as data — never instructions — when it reaches the agent.
- Sharper triage: the agent now uses the page, component, and app version to classify better, and flags anything ambiguous or risky as “Needs review” so it waits for you instead of auto-opening a PR.
- New docs: a direct Feedback API reference and a Public board guide.
v0.8.0
A more trustworthy, more polished worker
June 2026- Verify before opening a PR: point the worker at your repo's tests/build (`"verify": true` in worker.json) and it runs them in the isolated worktree first. A fix that fails still opens a draft PR — clearly flagged — but is never auto-merged.
- Auto-merge got stricter: it now refuses changes to CI, Dockerfiles, lockfiles, and shell scripts, and won't merge anything that failed verification.
- Lost work is recovered automatically. If your laptop sleeps or the worker dies mid-job, a reaper requeues the job within minutes instead of leaving it stuck forever.
- The dashboard now shows when the worker is actively Working (not just online), warns if a job looks stuck, and displays the worker version + machine name.
- Hardened: your feedback text is handled as data the agent reads, never as instructions it could be tricked into following; the worker warns if your token file is readable by others.
- The worker CLI itself got a glow-up — a proper banner, color-coded live status, and a readable run-by-run log as it triages and fixes.
v0.7.0
Daily summary email
June 2026- Turn on a once-a-day digest in Account → Notifications: PRs your worker opened, anything that failed, and what's awaiting your review — so you can review on your own schedule without watching the dashboard.
- Opt-in and off by default. Quiet days are skipped (no empty emails).
v0.6.0
Open the widget from any element
June 2026- The floating button is now just the default. Add `data-fdbck` to any element — a branded button, a footer “Support” link, a menu item — and it opens the same feedback modal. No JavaScript, and it works for elements rendered after load.
- `data-fdbck-type="bug|feature|other"` lets a trigger preselect the feedback type, so a “Report a bug” link lands users on the right form.
- New JS API for framework apps: `fdbck.open()`, `fdbck.close()`, `fdbck.toggle()` — `open()` takes an optional `{ type }`.
- Going launcher-less (great on mobile)? Set `data-hide-fab="true"` on the script tag to drop the floating button and drive the modal entirely from your own UI.
v0.5.0
Multi-vendor agents
June 2026- The worker now runs whichever agent CLI you've installed — Claude Code, OpenAI Codex, or Google Gemini — not just Claude.
- Pick the vendor and model per phase (triage vs. code) in Account → Models. Still zero keys stored on fdbck: it all runs on your own subscription.
- Power users can override the exact command per vendor in worker.json via a `commands` map.
- Every PR fdbck opens now carries a brand-orange `fdbck` label, so they're easy to filter in your repo's Pull requests.
v0.4.0
Onboarding & docs
June 2026- First-run setup checklist in the inbox — embed the widget, run the worker, send a test — each step lights up green the moment it's done.
- Live local-worker status: a pill in Account shows whether your worker is online, idle, or offline.
- Public docs at /docs: getting started, the full widget embed reference, running the worker, automation, and troubleshooting.
- “Send test feedback” fires a synthetic item through the whole pipeline so you can prove the loop without waiting for real traffic.
- Auto-PR and Auto-merge now use plain-English presets (Conservative / Balanced / Aggressive) with the raw thresholds tucked under Advanced.
- fdbck-worker doctor — one command validates your whole setup and tells you exactly what's missing.
- Inline help tooltips and Docs links throughout settings; regenerate your widget key in one click.
v0.3.0
Zero server keys — your Claude, your machine
June 2026- fdbck no longer stores any model keys. Triage and fixes both run on your own machine via the fdbck-worker CLI and your own Claude Code.
- The local worker claims feedback, triages it, writes the fix in an isolated git worktree, and opens a draft PR — all with your own credentials.
- Per-phase model preferences: pick a fast model for triage and a stronger one for code.
v0.1.0
The core loop
Earlier- Embeddable feedback widget with automatic, PII-scrubbed page-context capture.
- Triage that classifies type, severity, and confidence, then gates which items get an automatic fix.
- Draft pull requests against an isolated branch, with optional auto-merge for high-confidence bug fixes.
- Multi-app support, per-project target branch and monorepo root, and a trust dashboard tracking clean-merge rate.