Templates from the desk — not the internet.

01

Product requirement sketch

When I use it: before writing APIs for a desk workflow — not a 40-page PRD.

• Who operates this daily? Who decides? Who gets blamed when it fails?
• What is the current workaround (spreadsheet, memory, inbox)?
• What does “done” look like on a Monday morning?
• Which errors are embarrassing vs career-threatening?
• Domain objects (e.g. company → file → matter) — not feature bullets.
• Out of scope for v1 — explicitly listed.
• Success signal you can observe in production (not vanity).

How I build · NepalIPMS case

02

AI feature evaluation checklist

When I use it: before adding OCR, drafts, or “assist” to a system of record.

• Does this remove repeated work — or only impress in a demo?
• Where does the suggestion land? (Must be a review surface.)
• Can a wrong accept create silent liability?
• Is confidence visible beside every field/suggestion?
• What is the human override path?
• How will we measure usefulness after ship (time saved, rework, trust)?
• Is the model the product — or a stage in a workflow?

Playbook · HITL pattern

03

Product discovery checklist

When I use it: sitting with a desk before designing screens.

• Watch one full real task end-to-end (not a interview summary).
• List every tool touched (drive, mail, sheet, WhatsApp, paper).
• Capture the moment trust breaks (wrong mark, missed deadline).
• Map who must confirm before the system of record updates.
• Note bilingual / messy input realities.
• Write the friction sentence in their words.

04

Engineering handoff template

When I use it: when another builder (or future me) must operate the slice.

• Purpose of this slice in one sentence.
• Domain objects touched; migrations required.
• API surface + auth assumptions.
• Failure modes and how they surface to the desk.
• How to deploy; how to roll back.
• What is intentionally unfinished.
• Link to decision / ADR if a tradeoff was locked.

05

Architecture review checklist

• Does the data model match how the desk already works?
• Where is tenant isolation enforced?
• What happens when OCR/AI is wrong?
• Can one person operate this stack?
• Expand-contract path for schema change?
• Observability: what do we log without PII?
• Diagram exists and is downloadable?

Diagrams

06

UX review checklist (non-technical operators)

• Can the task be completed without reading docs?
• Is the next action obvious after every save?
• Are dangerous actions reversible or confirmed?
• Does search return something trustworthy under time pressure?
• Are empty states honest (not marketing)?
• Mobile/desk: which device is primary for this role?

07

Release checklist

• Migrations applied (or expand step live).
• Feature flag / path for rollback named.
• Desk-critical path smoke-tested with real-shaped data.
• Changelog / journal note drafted (what + why).
• No silent AI write paths introduced.
• Status / product page updated if user-facing.

08

Deployment checklist (Cloudflare-first)

• wrangler / CI deploy of Worker + assets.
• D1 migrations version-checked.
• R2 / KV bindings present in prod.
• Secrets not in client bundles.
• Smoke: health, auth, one write, one search.
• Confirm previous version ID known for rollback.

Cloudflare-first pattern

09

Technical decision record (ADR)

Mirror of the public decision log schema.

• Context — forces and constraints.
• Alternatives — at least two real options.
• Decision — what we chose.
• Trade-offs — what we give up.
• Outcome — what happened after ship.
• Same today? — revisit date or yes/no with reason.

Decision log

10

Incident postmortem template

Blameless. Desk impact first.

• Problem — what the desk experienced.
• What happened — timeline of facts.
• Why — contributing causes (not villains).
• What changed — code, process, monitoring.
• What we’d do differently — prevention.
• Customer/comms notes if external impact.

Lessons from production