Published items
Minimal AGENTS.md starter
A lean AGENTS.md template: project context, commands, conventions, and hard rules.
TDD Feature Loop
Multi-step agent workflow. A red-green-refactor loop for building one feature with a coding agent: one failing test at a time, minimal implementation, refactor under green, with stop points before scope changes.
Spec, Plan, Implement, Verify
Multi-step agent workflow. A four-gate workflow for medium-sized changes: write a spec the user approves, plan the files and steps, implement in reviewable chunks, and verify against the spec before declaring done.
Release Cut and Notes
Multi-step agent workflow. Cuts a release from the main branch with checks at each step: verify the tree, choose the version from the changes, generate notes from commits, tag, and hand off with a rollback plan.
Prompt Iteration With a Small Eval Set
Multi-step agent workflow. Improves a prompt empirically: build a 15-30 case eval set with a rubric, score the baseline, change one thing at a time, and keep only changes that move the score without regressing held-out cases.
Legacy Module Strangler Migration
Multi-step agent workflow. Replaces a legacy module incrementally: pin current behavior with tests, route calls through a facade, migrate one capability at a time with shadow comparison, retire on evidence.
Upgrade a Dependency Safely
Multi-step agent workflow. Upgrades one dependency with evidence at each step: read the changelog for breaking changes, upgrade in isolation, run the full verification, and document the risk for the reviewer.
Codebase Onboarding, Day One
Multi-step agent workflow. A structured first pass through an unfamiliar repository that ends with a verified local build, a map of the architecture, and a list of questions for the team, without changing any code.
Bug Report to Pull Request
Multi-step agent workflow. Takes a bug report to a mergeable PR: reproduce first, pin the root cause with a failing regression test, fix minimally, verify, and write a PR description a reviewer can trust.
Technical Blog Post Outline
Turns a topic and raw notes into a tight blog post outline with a one-sentence thesis, a hook grounded in a real problem, and section-level evidence requirements.
Release Notes From a Changelog
Converts raw commit messages or a changelog into user-facing release notes grouped by impact, with breaking changes and required actions called out first.
README From a Codebase
Writes a README grounded only in what the repository actually contains: real commands from the manifest, real entry points, and honest gaps marked instead of invented.
Executive Summary of a Technical Document
Distills a long technical document into a one-page executive summary that leads with the decision or ask, quantifies impact, and stays faithful to the source.
API Reference From Code
Generates reference documentation for a module's public surface directly from source: signatures, parameters, return values, errors, and a tested-looking example per symbol.
Threat Model a Feature
Produces a scoped STRIDE-style threat model for one feature: assets, trust boundaries, ranked threats with concrete attack paths, and mitigations mapped to each.
PR Review Against a Rubric
Reviews a pull request against an explicit, weighted rubric and returns line-anchored findings with a clear merge recommendation.
Incident Timeline Reconstruction
Rebuilds a precise, source-attributed incident timeline from chat logs, alerts, and deploy records, separating what happened from when it was noticed and flagging gaps.
Architecture Review of an ADR
Reviews an Architecture Decision Record for decision quality: are the alternatives real, the consequences honest, and the context sufficient for someone to revisit it later.
Accessibility Audit of a UI Component
Audits a UI component's markup and behavior against WCAG 2.2 AA, with each finding tied to a success criterion, an assistive-tech consequence, and a code-level fix.
Summarize a Long Transcript
Summarizes a long meeting, interview, or call transcript into decisions, action items with owners, open questions, and a faithful narrative, quoting speakers where it matters.
Structured Comparison of Options With Decision Matrix
Compares N libraries, tools, or approaches against weighted criteria you define, producing a sourced decision matrix and a recommendation that states what would change it.
Fact-Check With Citations
Checks each factual claim in a passage against supplied sources, returns a verdict per claim with quoted evidence, and refuses to rule on claims the sources cannot settle.
Extract Structured Data From a Document
Pulls fields from an unstructured document into JSON that conforms to a schema you supply, with nulls for absent data and a confidence note on anything inferred.
Self-consistency: arithmetic CoT exemplars (DAIR.AI)
Few-shot CoT exemplar set (Wang et al. 2022, Table 17) used to sample multiple reasoning paths and take the majority answer.
ReAct prompting: thought/action/observation exemplar (DAIR.AI)
Single ReAct-format trajectory exemplar (Yao et al. 2022, HotpotQA) from the Prompt Engineering Guide, for use as a few-shot demonstration in tool-using agents.
Prompt chaining for document QA: extract quotes, then answer (DAIR.AI)
Two-prompt chain from the Prompt Engineering Guide: prompt 1 pulls relevant quotes into <quotes> tags, prompt 2 composes the answer from quotes plus the original document.
Generated knowledge prompting: two-step commonsense QA (DAIR.AI)
Two-step prompt from the Prompt Engineering Guide (Liu et al. 2022): first have the model generate relevant facts, then answer using those facts.
Few-shot prompting: new-word usage demo (DAIR.AI)
One-shot/few-shot demonstration prompt from the Prompt Engineering Guide: show a worked example, then leave the final slot for the model to complete.
Chain-of-thought prompting: few-shot and zero-shot (DAIR.AI)
Few-shot chain-of-thought exemplars (Wei et al. 2022) plus the zero-shot 'Let's think step by step' trigger (Kojima et al. 2022) from the Prompt Engineering Guide.
Competitive Landscape Brief
Builds a competitive landscape brief from supplied research notes: positioning map, per-competitor profiles with sourced claims, and the gaps your product could own.
SLO Definition Workshop
Facilitates defining SLOs for a service: picks user-facing SLIs, proposes targets grounded in current performance, sizes the error budget, and specifies the alerting policy.
Runbook From an Incident
Turns a resolved incident's postmortem and chat log into an operational runbook: detection signals, exact diagnostic commands, decision points, and the mitigation that actually worked.
On-Call Handoff Summary
Compresses a shift's alerts, tickets, and chat into a handoff the incoming on-call can act on in five minutes: active issues first, with current state, next step, and tripwires.
Zero-Downtime Migration Plan
Plans a zero-downtime migration (schema, datastore, service, or provider) as reversible phases with verification gates, dual-write or dual-read strategy, and explicit rollback at every step.
Cloud Bill Cost Optimization Review
Analyzes a cloud cost export to find the largest savings by effort level, with the exact resources, estimated monthly impact, and the risk of each change stated plainly.
Write SQL From a Schema
Turns a plain-English data question into a correct, dialect-specific SQL query using only the tables and columns in the supplied schema, with join logic and assumptions made explicit.
Write the Failing Test First
Turns a behavior description into one focused failing test in the project's existing test style, before any implementation code is written.
Regex With Tests
Builds a regular expression for a stated matching rule in a specific engine, explains it piece by piece, and ships a table of positive and negative test cases.
Refactor Within Explicit Constraints
Refactors code toward a stated goal while holding hard constraints (public API, behavior, dependencies) fixed and proving each one still holds.
Generate a Typed API Client From OpenAPI
Produces a hand-quality typed client from an OpenAPI spec: one function per operation, shared types, explicit error handling, and no runtime dependency surprises.
Explain Unfamiliar Code
Walks through a block of unfamiliar code top-down, grounding every claim in quoted lines and flagging what cannot be known from the snippet alone.
Design Doc From Requirements
Expands a requirements list into a reviewable engineering design doc with explicit non-goals, two or more considered alternatives, and a rollout plan.
Debug From a Stack Trace
Reads a stack trace and the relevant source, forms ranked hypotheses with evidence, and proposes the single cheapest experiment to confirm the root cause.
Convert Code Between Languages
Ports code from one language to another idiomatically, preserving behavior exactly and documenting every place the two languages' semantics diverge.
Pre-PR Self-Review Checklist
Reviews your own diff the way a demanding reviewer would, before you open the PR: correctness, tests, scope creep, and the questions a reviewer will ask.
Technical Documentation Writer
System prompt. System prompt for a documentation writer that produces accurate, task-oriented docs from source material, follows a house style guide, and marks every unverifiable claim instead of guessing.
Read-Only SQL Analyst
System prompt. System prompt for a data analyst agent that answers business questions by writing and running read-only SQL, states its assumptions, and never guesses at columns it has not seen.
Socratic Programming Tutor
System prompt. System prompt for a programming tutor that leads learners to their own answers through targeted questions and small experiments, giving the solution only when the learner is truly stuck.
Research Assistant With Citations
System prompt. System prompt for a research assistant that answers from retrieved sources, cites every claim inline, separates evidence from inference, and reports what it could not find.
Product Spec Writer
System prompt. System prompt for a product spec writer that turns rough ideas and research into PRDs with measurable success criteria, explicit non-goals, and open questions instead of invented certainty.
Internal Knowledge Base Q&A (RAG Answerer)
System prompt. System prompt for a retrieval-augmented answerer over an internal knowledge base: answers only from retrieved chunks, cites them by ID, and says 'not in the knowledge base' when the answer is absent.
Incident Commander Copilot
System prompt. System prompt for an assistant that supports a human incident commander: tracks state, drafts status updates, proposes next actions with risk, and never executes changes itself.
Data Extraction Agent (Strict JSON Only)
System prompt. System prompt for an extraction agent inside a pipeline: reads input documents, emits one JSON object that validates against a given schema, and signals failure in-band instead of writing prose.
Customer Support Agent for a SaaS Product
System prompt. System prompt for a tier-1 SaaS support agent that resolves issues from documented knowledge, uses account tools carefully, and escalates with a complete handoff when it cannot resolve.
Code Review Bot for GitHub Pull Requests
System prompt. System prompt for an automated PR reviewer that posts a small number of high-confidence, line-anchored comments on correctness and security, and stays silent on style unless asked.
UX/UI developer persona (awesome-chatgpt-prompts)
System prompt. Persona prompt: propose user-experience improvements, prototypes, and design tests for a described app or site.
Tech writer persona (awesome-chatgpt-prompts)
System prompt. Persona prompt: turn a list of basic app steps into an engaging how-to article, marking where screenshots belong.
Cyber security specialist persona (awesome-chatgpt-prompts)
System prompt. Persona prompt: given how data is stored and shared, propose protection strategies such as encryption, firewalls, and suspicious-activity policies.
Code review assistant persona (awesome-chatgpt-prompts)
System prompt. Persona prompt: structured code review covering readability, bugs, performance, and standards, with a fixed three-part response format.
TODO and FIXME sweep
Claude Code slash command (.claude/commands/todo-sweep.md). Finds every TODO, FIXME, HACK, and XXX comment, dates and attributes each via git blame, triages them into fix-now / ticket / delete, and offers to act.
Collections
- Prompt essentials (5 items)
- Data science pack (5 items)
- Python services pack (6 items)
- React & TypeScript pack (6 items)
- Agent starter pack (4 items)