What does an optimized coding prompt carry that a one-line request does not?

The role to adopt, the artefact to produce, the input and output shape, the edge cases the request implies, and what to do when the code is ambiguous. Text prompts come back as XML with named sections — role, task, requirements, output_format, verification — so the model receives the shape of the answer rather than only its subject.

Will a prompt here assume a language or toolchain I do not use?

It should not. The generator is explicitly barred from inventing technical constraints — a shell version, a POSIX-only rule, a banned tool — unless the original request stated them. Anything environment-specific in a listing came from its author, so read it as an example and swap it for your own stack.

How do I run one of these?

Copying adds it to your own history, which needs an account; from there you can edit it and run it against any model your plan allows. Reading one works too: the constraints it spells out are usually the ones missing from the prompt you would have typed.

All prompts

Coding prompts

Prompts for writing, reviewing and repairing software — each one already expanded into an explicit brief.

A one-line request like "make this faster" leaves a model guessing at the language, the constraint that matters and what evidence you want back. These prompts were run through the optimizer, so they carry the parts that decide the answer: the role to adopt, the artefact to produce, whether to show a diff or a rewrite, and what to do when the code is ambiguous. Copy one into your own history to edit it against your stack.

7 prompts published

Promptcoding

Zero-Downtime Production Deployment Runbook

A deployment plan that assumes something will go wrong. Starts with the prerequisites people discover too late — secrets injection, environment isolation, database migrations — then compares Blue/Green, Canary and Rolling against your architecture instead of picking for you. Defines health-check endpoints, the metrics and error budgets to watch, and the alert thresholds that fire. Every step names its rollback trigger, so the abort decision is made before the deploy.

55 0 0by Andrei Badulescu

Preview

Promptcoding

Feature Architecture Review — Trade-Off Matrix Before Any Code

For the decision you make before you open an editor. Puts two or three viable approaches side by side — modular monolith, event-driven, serverless — in a matrix of pros, cons and the scenario each actually suits, then commits to one and details its data flow, boundary contracts and failure handling. Also defines how the feature gets shipped and kept alive: pre-commit checks, CI stages, a review checklist written against its failure modes. Asks first when context is thin.

86 0 0by Andrei Badulescu

Preview

Promptcoding

PostgreSQL Schema + Reversible Migration (Up and Down)

Produces the ER structure and the DDL together, so the design and the thing you actually run cannot drift. Normalized to 3NF, identity or UUID primary keys, explicit ON DELETE and ON UPDATE on every foreign key, TIMESTAMPTZ timestamps, CHECK constraints on bounded values, B-tree indexes on foreign keys and high-cardinality filters — and a warning against indexing low-cardinality flags. Both migrations are wrapped in transactions, and the down script drops in reverse dependency order.

63 0 0by Andrei Badulescu

Preview

Promptcoding

Performance Refactor with Before/After Complexity Analysis

Cleanup with a contract: external inputs, return values and API surface stay exactly as they were, so the diff is safe to merge without touching callers. Flattens nested conditionals into guard clauses, renames toward self-documenting identifiers, removes dead code, commented-out blocks and orphaned imports, and replaces silent catches with explicit handling. Comments are added only where domain logic genuinely needs them. Ends with a list of what changed and why, so review is fast.

71 0 0by Andrei Badulescu

Preview

Promptcoding

Production Incident Fix — Root Cause, Minimal Patch, Rollback Plan

Written for the hour something is actually down. Triages the failure mechanism first — uncaught exception, race, resource leak, serialization mismatch, deadlock — states the root cause in two or three sentences, then patches. Scope is held deliberately tight: minimal viable fix, no refactoring, no API or schema breaks. Comes back with the changed file, a repro test that fails before and passes after, and the metrics and rollback plan to watch on deploy. Paste the stack trace or the failing code.

79 0 0by Andrei Badulescu

Preview

Promptcoding

Codebase Walkthrough + Severity-Rated Technical Audit

Does the two jobs you actually need when you inherit a repo: explains the architecture, execution paths and domain models, then audits them. Findings come back in a table rated Critical through Low, each with location, failure mode and a remediation step — plus the guardrails that stop the same class of bug returning: pre-commit hooks, CI stages, a review checklist, onboarding notes. Bans vague critique: every finding must name a file, function or pattern. Paste a repo tree or a module.

101 0 0by Andrei Badulescu

Preview

Promptcoding

Multi-Tenant SaaS MVP Backend — Prisma, JWT, Query-Level Tenant Isolation

Turns a one-line product idea into a production-grade backend scaffold: a Prisma schema with Tenant, User, Membership and a tenant-scoped domain model; JWT middleware that extracts tenant context per request; RBAC for Admin and Member; centralized error handling with typed exceptions. Isolation is enforced in the query layer, not the application layer. Output arrives split by file path — schema, middleware, service, controller. Replace the stack line if you are not on TypeScript and Postgres.

97 0 0by Andrei Badulescu

Preview

Guides on coding prompts

Prompts That Return Code and Nothing Else

Code output has one hard requirement — it must run. Pinning the version, the imports, the file boundary, and getting the explanation out of the way without losing the assumptions.

Nine Lazy Prompts, Rebuilt

Nine one-line developer prompts run through the optimizer, with the verbatim output for each and the constraints none of them asked for.

Questions about coding prompts

What does an optimized coding prompt carry that a one-line request does not?: The role to adopt, the artefact to produce, the input and output shape, the edge cases the request implies, and what to do when the code is ambiguous. Text prompts come back as XML with named sections — role, task, requirements, output_format, verification — so the model receives the shape of the answer rather than only its subject.
Will a prompt here assume a language or toolchain I do not use?: It should not. The generator is explicitly barred from inventing technical constraints — a shell version, a POSIX-only rule, a banned tool — unless the original request stated them. Anything environment-specific in a listing came from its author, so read it as an example and swap it for your own stack.
How do I run one of these?: Copying adds it to your own history, which needs an account; from there you can edit it and run it against any model your plan allows. Reading one works too: the constraints it spells out are usually the ones missing from the prompt you would have typed.