2026-03-20 — Prakta

What shipped today

Massive session — 27 issues shipped, a complete product vision pivot, and the foundation for the serve-don’t-show model laid end to end.

UX pivot: “serve, don’t show.” The biggest outcome wasn’t code — it was the brainstorm that fundamentally reshaped what Prakta is. Through an extended conversation, we landed on: the user never sees a task list. Prakta serves one chunk at a time. Check out / check in, like Subversion. Energy as a number (not just 4 states). One cycle length per user. Decomposition at ingestion. No priorities — energy + mode + grain chain only. The PRODUCT.md rewrite captures the full model with all edge cases resolved: partial check-ins, empty queues, multi-day tasks, the escape hatch problem, “not now” spirals (not a supervisor), and more.

Foundation layer shipped. Every schema change, config update, and core function the serve model requires is now in place: single cycle length (#133), cycle-sized decomposition at parse time (#134), energy_cost on tasks (#157), numeric energy budget at day-start (#158), mode preference to start the grain chain (#140), started_at timestamps (#153), checkin_notes for feedback (#163), source task completion rollup (#137), and the serve API itself (#151) — a deterministic scoring function that picks the best chunk without calling Claude.

UX changes landed. Task list hidden by default with /queue escape hatch (#136). Narrative queue via Maya “What’s ahead?” button (#142). Daily cap enforcement — “you’re done” when cap is hit (#146). Weekly reflection report (#143). Homepage copy rewritten for the serve model (#148). Deadline_weight removed from sequencing — pure no-priority model (#149). Onboarding redirects to /today (#147).

Earlier in the session: production 500 fixed (#111), app nav bar (#112), sign-out button (#117), screenshot ingestion via Claude Vision (#124), task edit/delete + dashboard list (#125/#128), input length limits (#113), JSON error handling (#115), sequencing validation (#116), daily plan persistence (#114), learning insights after wrap-up (#126), task history page (#127).

Completed

• #111 — Fix production 500 on www.prakta.io
• #112 — App navigation bar
• #113 — Input length limits on AI API routes
• #114 — Persist daily plan in database
• #115 — request.json() error handling
• #116 — Validate sequencing AI response
• #117 — Sign-out button
• #124 — Screenshot task ingestion (Claude Vision)
• #125 — Task edit and delete
• #126 — Learning model insights after wrap-up
• #127 — Task history page
• #128 — Show task list on dashboard (included in #125)
• #133 — Single cycle length (collapse 4 mode-specific)
• #134 — Decompose at ingestion to cycle size
• #136 — Hide task list by default
• #137 — Source task completion rollup
• #140 — Day-start mode preference
• #142 — Narrative queue view via Maya
• #143 — Weekly reflection report
• #146 — Daily cap enforcement
• #147 — Onboarding redirect to /today
• #148 — Homepage copy update
• #149 — Remove deadline_weight (no-priority model)
• #151 — Serve API (deterministic pick-one-chunk)
• #153 — started_at timestamp tracking
• #157 — energy_cost field on tasks
• #158 — Numeric energy budget at day-start
• #163 — checkin_notes field for feedback

Decomposed: #135 (serve experience → #151, #152, #153), #139 (energy budget → #157, #158, #159), #141 (check-in feedback → #163, #164, #165)

Release progress

March 2026: 0 open / 0 closed (milestone exists, no issues assigned)

Carry-over

• #152 — Serve UI: The last critical piece — wire the serve API into the /today page. Replaces multi-cycle plan view with one-chunk-at-a-time. Needs prep.
• #159 — Energy tracking: Track energy_spent when tasks complete, stop serving when budget exhausted.
• #164 — Feedback UI: Add text input to done/not-done flow (checkin_notes schema ready).
• #165 — Re-ingestion pipeline: Not-done feedback re-ingested through parse.
• #101 — Email templates: Still waiting for Paul to paste into Supabase dashboard.
• Google OAuth + Stripe: Manual setup still pending.

Risks

• No tests. 50+ PRs merged with zero test coverage. The first real user will be the integration test.
• Serve UI (#152) not built yet. The serve API exists but the /today page still shows the old multi-cycle plan view. The serve model is backend-complete but UI-incomplete.
• Production not redeployed since morning’s #111 fix. All of today’s work is on main but not deployed.

Flags and watch-outs

• Deploy from monorepo root, not web/. Project: synaxis/prakta.
• 6 new Supabase migrations today (plan_json, learning_insight, started_at, energy_cost, energy_budget+spent, checkin_notes). All applied to remote.
• buttonVariants server/client split — shadcn overwrite clobbers it.
• Port 3006 always.
• Serve API is deterministic — no Claude call. Decision logged in DECISIONS.md.

Next session

1. /issue prep 152 --auto then /issue exec 152 --auto — build the serve UI (the last critical piece)
2. Deploy to production: vercel --prod --yes from monorepo root
3. Test the full production flow: signup → onboarding → tasks → serve → wrap-up
4. Prep and exec #159 (energy tracking)
5. Prep and exec #164 (feedback UI) and #165 (re-ingestion)
6. Google OAuth + Stripe manual setup
7. Consider assigning issues to the March 2026 milestone