Prompts for AI coding agents

238 prompts, commands and recipes. Each one says what it returns and when to skip it. Copy one and paste it into your agent.

Engineering investigations

Prompts that send an agent looking for one kind of problem without changing anything. Each says what proof to bring back and when to skip it.

Architecture and hidden complexitySpots where the way the code is built invites bugs: status spread across many places, words that change meaning between layers, and layers that cost more than they help.8 entries Performance and scaleWhere time and money go as the data grows. Every claim here needs a measurement before and after.9 entries Correctness and reliabilityWhat happens when something runs twice, fails, times out or happens in the wrong order, and whether people can tell a stand-in value from the real one.7 entries Testing and securityWhat the tests really prove, and where input from outside can reach something sensitive.7 entries Infrastructure and environmentThe path from a merged change to the live site, what automated checks cost, and the computer the agents run on.7 entries Data integrity and analyticsTelling missing from zero, knowing where each number came from, and records whose meaning changes over time.6 entries Agent harness and modern stackNewer kinds of trouble: out-of-date agent instructions, old and new versions running at once after a deploy, data leaking into pages, outside services going down, rounding errors, and screens that show a change before it is saved.6 entries

Database

Prompts for PostgreSQL databases on any host: proving the data is right, simpler and cheaper writes, a leaner schema, real costs, and changes tested on a copy. All read-only unless you say otherwise.

Start hereOne broad look when the database feels hard to trust or costly. If you can already name the problem, skip it and pick the narrower prompt.1 entry Accuracy and evidenceProves the data is right, not just valid: rebuilt figures, whole populations, stable identities and past results that still reproduce.4 entries Writes and ingestionWho writes what, how often it changes nothing, and whether imports survive retries and corrections.5 entries Schema and dependenciesFewer objects, one home for each fact, and a model a new engineer or agent can read.3 entries Performance and maintenanceWhere storage, time and write cost actually go, and what grows expensive first. Every claim needs a measurement.2 entries Rebuild and safe changeChanges and restores proved on a separate database before anyone relies on them.2 entries Agent decisionsTurns questions agents keep asking you into rules they can apply themselves.1 entry WorkflowsShort routes through the database prompts for a whole job. Start with the first step and only take the optional ones when the evidence points there.5 entries

Thinking prompts

Short lines you add to any task to change how the agent works: try to prove itself wrong, keep things simple, check its work, hand off cleanly.

Thinking prompts11 entries

Engineering commands

Installed skills you start with a slash command. Each one runs part of an investigation for you.

Engineering kit12 entries Trail of Bits11 entries Experience architecture3 entries Other commands that help8 entries

Design commands

Every design skill command, with a ready example prompt.

Impeccable27 entries Routers and working-style3 entries Built into Claude1 entry Wondelai UX design11 entries Color skills2 entries UX writing1 entry Marketing copy3 entries SwiftUI Pro1 entry Interface Design4 entries Hallmark4 entries Emil Kowalski skills15 entries Web Design Guidelines1 entry Anthropic example-skills8 entries Fix writing1 entry Humanizer1 entry Vale6 entries Stop Slop1 entry Sepia6 entries slop-guard2 entries

Design recipes

Short sequences of design commands for a whole job, from a new landing page to dark mode.

Design recipes22 entries