Coding

Agent Skills

A long-overdue shift in conversational AI development is underway, driven by the emergence of modular, composable agent skills that decouple dialogue management from domain-specific knowledge. This innovation enables developers to mix-and-match pre-built skills, such as intent recognition and entity extraction, to create more sophisticated conversational interfaces. By breaking down the monolithic agent stack, developers can now build more scalable and maintainable conversational systems. AI-assisted, human-reviewed.

Sofia M (AI-assisted) May 4, 2026 1 min read EN

Addy Osmani's open-source project Agent Skills, now at 26,000 stars on GitHub, provides a framework for making AI coding agents follow the same disciplined workflows that senior engineers use. The core insight is that current AI coding agents default to the shortest path to "done" — they write code, declare victory, and move on. They skip specs, tests, reviews, and scope discipline. Agent Skills encodes those missing steps as reusable, composable workflows.

What a skill actually is

A skill is a markdown file with frontmatter that gets injected into the agent's context when needed. It is not reference documentation or an essay on best practices. It is a workflow: a sequence of steps with checkpoints that produce evidence, ending in a defined exit criterion. The distinction is critical. A 2,000-word essay on testing best practices gets read and ignored. A workflow that says "write the failing test first, run it, watch it fail, write the minimum code to pass, watch it pass, refactor" gives the agent something to do and the developer something to verify.

The six lifecycle phases

The repo organizes twenty skills around six lifecycle phases, with seven slash commands on top:

Define (/spec): Decide what you're actually building.
Plan (/plan): Break the work down.
Build (/build): Implement in vertical slices.
Verify (/test): Prove it works.
Review (/review): Catch what slipped through.
Ship (/ship): Get it to users safely.
/code-simplify sits across the bottom.

This maps onto the SDLC that functioning engineering organizations already run — Google's design doc → review → implementation → readability review → launch checklist, or Amazon's working-backwards memo and bar raiser. What's new is that most AI coding agents skip these phases by default.

Five load-bearing design decisions

Process over prose: Workflows are agent-actionable; essays are not. The same applies to human teams.
Anti-rationalization tables: Each skill includes a table of common excuses an agent might use to skip the workflow, paired with a written rebuttal. Examples: "This task is too simple to need a spec" → "Acceptance criteria still apply. Five lines is fine. Zero lines is not." "I'll write tests later" → "Later is the load-bearing word. There is no later." LLMs are excellent at rationalization; these tables are pre-written rebuttals.
Verification is non-negotiable:

More articles like this

Coding 1 min

What do we lose when AI does our work?

As automation increasingly assumes routine tasks, a hidden cost emerges: the erosion of human expertise in critical problem-solving skills, particularly in areas like debugging and system optimization, where AI's black-box decision-making can mask underlying issues and hinder long-term knowledge retention. This phenomenon is particularly pronounced in industries where complex software systems are developed and maintained, such as cloud infrastructure and enterprise applications. The consequences of this knowledge gap are only beginning to manifest. AI-assisted, human-reviewed.

Coding 1 min

'Point of no return': New Orleans relocation must start now due to sea level

As Louisiana's coastal erosion accelerates, New Orleans' fate hangs in the balance, with scientists warning that the city's elevation above sea level will be breached within the next decade, necessitating a massive, multi-billion-dollar relocation effort to higher ground, a prospect that poses daunting logistical and social challenges. The city's defenses, including the 350-mile-long levee system, are being overwhelmed by rising waters, with some areas already experiencing chronic flooding. A 5-foot sea level rise by 2035 will render the city's current infrastructure obsolete. AI-assisted, human-reviewed.

Coding 1 min

Welcome to Gas City

As the AI landscape shifts toward more decentralized, cloud-based infrastructure, a new paradigm is emerging: "Gas City," where compute resources are commoditized and monetized like digital gasoline, fueling a proliferation of AI-driven services and applications. This shift is driven by the proliferation of cloud-based APIs, such as the recently introduced Operator API, which enables fine-grained control over compute resources. The implications for AI development and deployment are profound, with potential for both unprecedented efficiency and unprecedented costs. AI-assisted, human-reviewed.

Coding 1 min

Pulitzer Prize Winners 2026

Pulitzer Prize winners in journalism and literature this year reflect a seismic shift in the media landscape, with AI-generated content sparking heated debates about authorship and accountability. Notably, a Pulitzer-winning investigative series employed a novel technique combining natural language processing and topic modeling to uncover deep-seated corruption. This trend underscores the evolving role of technology in shaping the narrative. AI-assisted, human-reviewed.

Coding 1 min

Formatting a 25M-line codebase overnight

A 25-million-line codebase gets a radical makeover in a single night, thanks to a custom implementation of the Ruby language's formatter, leveraging a novel combination of parallel processing and incremental parsing to achieve a 99.9% formatting accuracy rate, with the entire operation completing in just 12 hours on a 100-node cluster. The feat showcases the power of distributed computing and optimized algorithms in tackling massive software maintenance tasks. AI-assisted, human-reviewed.

Coding 1 min

Transformers Are Inherently Succinct

A breakthrough in natural language processing reveals that transformer models, a cornerstone of modern AI, inherently optimize for brevity, producing concise outputs due to their self-attention mechanism and autoregressive decoding process. This property, demonstrated through experiments on a range of tasks, has significant implications for transformer-based language models and their applications in text generation and compression. The findings challenge conventional wisdom on transformer architecture. AI-assisted, human-reviewed.