agentleFS
Sign inStart free
VictoriakaeyVictoriakaey/build-reliable-agents/SKILL.md

build-reliable-agents

Skill0 starsChanged 7 months agoMITView on GitHub

A collection of engineering skills for building reliable LLM-based systems and full-stack AI applications. Use when starting a new AI project, designing an agent pipeline, writing or diagnosing prompts, exploring solutions to a technical problem, making prompt changes, debugging non-deterministic behavior, designing a database, selecting a model, deploying to production, reviewing code, or connecting an agent to a full-stack application.

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Every change has a before/after measurement

    line 45

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

build-reliable-agents

Overview

This skill collection captures patterns battle-tested in production LLM system development. Each sub-skill is a self-contained process for one recurring problem. They can be used independently or in combination.

When to Use Which Sub-Skill

SituationSub-Skill
Have an idea but don't know where to startai-system-design
Facing a problem with multiple possible solutionsproblem-exploration
Previous approach failed, need to reconsiderproblem-exploration
Deciding agent architecture (single vs multi, tool-use vs nodes)agent-architecture
Writing a new prompt from scratchprompt-design (Mode 1)
Existing prompt producing unstable or wrong outputsprompt-design (Mode 2)
Starting any implementation taskexperiment-driven-development
About to change a promptprompt-change-management
Need to compare two system versionsregression-testing
Designing a Critic, Judge, or Evaluator componentcritic-judge-design
Connecting agent to a web app, API, or third-party platformagent-integration
Designing or improving agent tools, observations, and evaluation signalsharness-design
Agent picks wrong tool, ignores data, or loops without progressharness-design (diagnosis ladder)
Designing a database schemadatabase-design
Choosing a model or deciding API vs localmodel-selection
Designing how an agent remembers informationmemory-system
Deploying to production or setting up CI/CDdevops
Reviewing code — Claude reviews or guided self-reviewcode-review
Starting a new LLM project from scratchai-system-design → agent-architecture → model-selection → database-design → experiment-driven-development
System behaving inconsistently across runsregression-testing → prompt-design (Mode 2)
Critic/Judge producing unstable judgmentscritic-judge-design → prompt-design (Mode 2)
Agent works locally but fails when integratedagent-integration → devops
Agent misbehaves — wrong tools, missed data, silent failuresharness-design → experiment-driven-development

Core Principle Across All Sub-Skills

LLM behavior is non-deterministic. Treat every change as an experiment, not a fix.

This means:

  • Every change has a before/after measurement
  • Regressions are expected and must be detected early
  • Prompt changes have blast radius beyond the component they touch
  • Deterministic components (validators, routers, formatters) are always preferable to LLM components when both can do the job

Sub-Skills

  • ai-system-design/SKILL.md — Go from a vague idea to a full system design, guided by conversation
  • problem-exploration/SKILL.md — Exhaust the solution space before committing to any approach
  • agent-architecture/SKILL.md — Agent architecture decision framework
  • prompt-design/SKILL.md — Write or diagnose prompts for any LLM component
  • experiment-driven-development/SKILL.md — 10-step process from evidence to git commit
  • prompt-change-management/SKILL.md — Safe prompt iteration with regression detection
  • regression-testing/SKILL.md — Controlled comparison protocol for LLM systems
  • critic-judge-design/SKILL.md — Designing reliable LLM-as-Judge components
  • agent-integration/SKILL.md — Connect agents to full-stack apps, APIs, and third-party platforms
  • database-design/SKILL.md — Schema design for general and AI-specific systems
  • model-selection/SKILL.md — API vs local, model selection, fine-tuning decisions
  • memory-system/SKILL.md — Design how agents remember and retrieve information
  • devops/SKILL.md — Deploy to production, CI/CD, monitoring, rollback
  • harness-design/SKILL.md — Design action space, tool boundaries, observation format, and evaluation signals
  • code-review/SKILL.md — Claude reviews your code or guides you through reviewing

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

Check your setup, free