DenisSergeevitch/agents-best-practices

Provider-neutral Agent Skill for Codex, Claude Code, and agentic harness design.

What it solves

This project provides a provider-neutral "Agent Skill" designed to help developers build, audit, and refactor agentic harnesses. It addresses the common problem of creating agents that are brittle, expensive, or unsafe by shifting the focus from prompt engineering to the rigorous design of the runtime environment (the harness) that validates, authorizes, and executes the model's proposed actions.

How it works

The project is implemented as a skill that can be installed into compatible AI agents (such as Codex or Claude Code). Once installed, the skill activates when a conversation discusses agent architecture, tool permissions, or production readiness. It leverages a comprehensive set of reference documents (found in the references/ directory) to guide the agent in generating MVP blueprints, auditing existing harnesses, and designing advanced features like recursive delegation or speculative tool execution.

Who it’s for

Developers and architects building agentic systems across various domains—including research, support, operations, sales, finance, and healthcare—who need a production-safe runtime discipline for their AI agents.

Highlights

  • MVP Blueprinting: Generates the smallest useful, production-safe agent harness based on a specific domain.
  • Harness Auditing: Identifies runtime-level failure points (e.g., missing loop budgets or unbounded tool results) rather than just prompt-level issues.
  • Permission Management: Guidance on splitting tools by risk class and implementing deterministic permission checks.
  • Advanced Architectures: Support for designing self-refining recursive harnesses, environment-adaptive tools, and always-on agents with durable sessions.
  • Performance Optimization: Strategies for prompt caching, speculative tool execution, and context memory compaction.
  • Hardware Integration: Contracts for installing agents on physical boards and performing hardware-specific evaluations.

Related

  • Project
    dzhng/skills

    A library of reusable, composable AI‑agent “skills” (prompt‑driven actions) for building autonomous software‑factory loops, installable via npm and usable with many coding assistants.

  • Project
    waynesutton/builder-skills

    builder‑skills provides a set of concise, markdown‑based “skills” that teach AI coding assistants how to follow Convex best‑practice rules and disciplined project‑workflow habits (PRDs, task tracking, safe git). Installable via skills.sh, Claude plugins, or npm, it drops starter docs and lets agents load the appropriate SKILL.md on demand, keeping code, docs, and safety checks in sync.

  • Project
    muratcankoylan/Agent-Skills-for-Context-Engineering

    A curated collection of “Agent Skills” that teach how to manage an LLM’s limited context window. The repo organises the material into self‑contained skill folders (foundational, architectural, operational, etc.) that can be loaded by Claude Code, Cursor, or any Open‑Plugins host. It explains context‑engineering concepts (lost‑in‑the‑middle, compression, self‑editable context), provides platform‑agnostic examples, and includes benchmark results showing high skill‑router accuracy on frontier models.

  • Project
    mxyhi/ok-skills

    A library of 31 ready‑to‑use “skills” (prompt‑plus‑script bundles) for AI coding agents like Codex, Claude Code, Cursor, OpenClaw and Autohand Code. Drop the repo into your `~/.agents/skills` (or Autohand) folder, merge the provided `CLAUDE_AGENTS.md` into your agent’s config, and then invoke skills such as `planning-with-files`, `find-docs` or `agent-browser` with natural language commands.

  • Project
    mgechev/skills-best-practices

    A comprehensive guide on best practices for creating structured agent skills to improve AI agent reliability, discoverability, and token efficiency.