{"slug": "the-next-programming-language-might-be-a-folder-of-skills", "title": "The Next Programming Language Might Be a Folder of Skills", "summary": "Anthropic's Agent Skills format became an open standard in December 2025, with Claude Code and Codex now supporting skills built around the portable SKILL.md folder shape, according to the article. A January 2026 paper found that a skill-based single-agent setup used about 54% fewer tokens and had about 50% lower latency on average while keeping comparable task accuracy on its controlled reasoning benchmarks and GPT-4o-mini setup. The paper also found that skill selection could fall short, and the figures are specific to those benchmarks rather than a general promise.", "body_md": "I used to think programming an agent meant two things:\n\nNow the interesting unit is often a folder.\n\nInside it: a SKILL.md, a few instructions, maybe a script, maybe some references. The agent discovers the folder, reads just enough to decide whether it matters, then loads the rest only when the task calls for it.\n\nAnthropic calls these Agent Skills [1]. The format became an open standard in December 2025 [2]. Claude Code and Codex now support skills built around that portable shape [3][4].\n\nNo, folders are not replacing Python. Put the pitchfork down.\n\nThe point is that we are gaining a new way to program **how an agent works**, not only what software it produces.\n\nAt minimum, an Agent Skill is a directory with one required file: SKILL.md. That file starts with YAML metadata and continues with the instructions the agent should follow [1][2].\n\nIt can also contain scripts, templates, examples, and longer reference material:\n\n```\npr-review/├── SKILL.md├── references/│   └── review-checklist.md└── scripts/    └── summarize-diff.sh\n```\n\nThat shape changes the relationship with the instructions. A pasted prompt disappears into chat history. A folder can live in git. I can diff it, review it, pin it, share it with a team, or delete it when it starts doing something strange.\n\nIn Claude Code, personal skills live under ~/.claude/skills/, while project skills live under .claude/skills/ and can travel with the repository [4].\n\n*Hand-drawn anatomy of a skill folder with SKILL.md, references, scripts, and assets*\n\nIf an agent loaded every installed skill in full at startup, a large library would eat the context window before I typed “hello.”\n\nSkills avoid that with **progressive disclosure** [1][2]:\n\nThink of it as a menu. The agent sees “PR review,” not the entire cookbook. If I ask it to review a patch, it opens that recipe. If the skill points to a security checklist, it reads the checklist at that moment.\n\nThis is why a skill feels different from another section in CLAUDE.md. Always-on instructions belong in the project context. A long workflow that matters twice a month belongs behind a skill name [4].\n\n*Hand-drawn three-step flow showing metadata, SKILL.md, and linked resources loading on demand* *Source: Image by the author, based on* *Anthropic’s Agent Skills explanation**.*\n\nA useful skill usually combines three kinds of material:\n\nThat third part matters. If the task is “sort these records,” the agent does not need to role-play a sorting algorithm one token at a time. It can run a script.\n\nThe folder becomes a bridge between fuzzy reasoning and boring code. Boring code is underrated. It tends to arrive at work on time.\n\nThese three ideas keep getting thrown into the same comparison, but they sit at different layers:\n\nAnthropic describes skills as complementary to MCP. A tool connection might let an agent read an issue tracker. A skill can teach the agent how your team triages those issues [1].\n\nClaude Code’s own guidance makes a similar distinction: use a skill for a reusable workflow in the main context; use a subagent when the work needs isolation, different tools, or a clean context window [5].\n\nMy shortcut:\n\n*Hand-drawn three-column guide showing skills for procedure, MCP for access, and subagents for isolation*\n\nA January 2026 paper asked whether some multi-agent workflows could be compiled into a single agent with a library of skills [6].\n\nOn its controlled reasoning benchmarks, the skill-based setup used about **54% fewer tokens** and had about **50% lower latency** on average while keeping comparable task accuracy [6]. Those figures belong to the paper’s selected benchmarks and GPT-4o-mini setup. They are not a promise for every coding project.\n\nStill, the design makes sense. If three agents only pass instructions to one another in sequence, a single agent may be able to load those behaviors as skills without paying for three separate conversations.\n\nThere is a catch.\n\nThe paper found that skill selection could fall sharply as libraries grew, especially when descriptions were semantically similar [6]. A folder full of api-helper, api-helper-pro, and better-api-helper-final is not a library. It is a cry for help.\n\nThe authors found hierarchical routing useful: choose a category first, then choose a skill inside it. Good naming and deletion matter too.\n\n*Hand-drawn comparison of a confusing flat skill list and a hierarchically routed skill library*\n\nSkills can contain scripts and instructions that use tools or reach the network. Installing one is closer to adding a package than bookmarking a prompt.\n\nAnthropic recommends using trusted sources and auditing a skill’s files, code, and external dependencies before use [1]. The open Agent Skills specification also gives us something valuable: a predictable directory to inspect [2].\n\nBefore I trust a skill, I want to know:\n\nThe scary skill is not always the obvious malware sample. Sometimes it is a cheerful 900-line SKILL.md that triggers on the word “code.”\n\nI would not start with a catalog of 100.\n\nPick one procedure you already repeat, such as PR review, and make the first version painfully small:\n\n```\n---name: pr-reviewdescription: Review a completed code diff for correctness, security, and API contract changes. Do not edit files.---\n```\n\nThen I would add:\n\nFinally, I would watch whether the agent loads it at the right time. A skill that never triggers is documentation. A skill that always triggers is a system prompt wearing a fake mustache.\n\nThe next programming language is not literally a directory tree.\n\nBut the folder is becoming a real unit of agent programming: metadata for discovery, Markdown for procedure, code for deterministic work, and references loaded only when needed.\n\nThat gives agent behavior some software-like properties we were missing. Skills can be versioned, reviewed, composed, shared, and audited. They can also conflict, bloat, and execute unsafe code, which is another very software-like property.\n\nSo I am done treating skills as fancy prompt snippets.\n\nThe model is the processor. The folder is starting to look like the program.\n\n[1] B. Zhang, K. Lazuka, and M. Murag, “Equipping agents for the real world with Agent Skills,” Anthropic Engineering, October 16, 2025. [https://www.anthropic.com/engineering/equipping-agents-for-the-real-world-with-agent-skills](https://www.anthropic.com/engineering/equipping-agents-for-the-real-world-with-agent-skills)\n\n[2] Agent Skills, “Specification.” [https://agentskills.io/specification](https://agentskills.io/specification)\n\n[3] OpenAI, “Skills in ChatGPT,” OpenAI Help Center, updated July 2026. [https://help.openai.com/en/articles/20001066](https://help.openai.com/en/articles/20001066)\n\n[4] Anthropic, “Extend Claude with skills,” Claude Code documentation. [https://code.claude.com/docs/en/skills](https://code.claude.com/docs/en/skills)\n\n[5] Anthropic, “Create custom subagents,” Claude Code documentation. [https://code.claude.com/docs/en/sub-agents](https://code.claude.com/docs/en/sub-agents)\n\n[6] X. Li, “When Single-Agent with Skills Replace Multi-Agent Systems and When They Fail,” arXiv:2601.04748, 2026. [https://arxiv.org/abs/2601.04748](https://arxiv.org/abs/2601.04748)\n\n[The Next Programming Language Might Be a Folder of Skills](https://pub.towardsai.net/the-next-programming-language-might-be-a-folder-of-skills-de884e732104) was originally published in [Towards AI](https://pub.towardsai.net) on Medium, where people are continuing the conversation by highlighting and responding to this story.", "url": "https://wpnews.pro/news/the-next-programming-language-might-be-a-folder-of-skills", "canonical_source": "https://pub.towardsai.net/the-next-programming-language-might-be-a-folder-of-skills-de884e732104?source=rss----98111c9905da---4", "published_at": "2026-09-21 13:31:01+00:00", "updated_at": "2026-09-21 13:53:43.822261+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "developer-tools", "agent-protocols"], "entities": ["Anthropic", "Agent Skills", "Claude Code", "Codex", "GPT-4o-mini"], "alternates": {"html": "https://wpnews.pro/news/the-next-programming-language-might-be-a-folder-of-skills", "markdown": "https://wpnews.pro/news/the-next-programming-language-might-be-a-folder-of-skills.md", "text": "https://wpnews.pro/news/the-next-programming-language-might-be-a-folder-of-skills.txt", "jsonld": "https://wpnews.pro/news/the-next-programming-language-might-be-a-folder-of-skills.jsonld"}}