OpenAI Quietly Ships Skills in ChatGPT and Codex CLI

OpenAI quietly shipped support for filesystem-based “skills” in both ChatGPT and Codex CLI, aligning with Anthropic’s lightweight approach. In ChatGPT, built-in skills guide robust document/PDF handling by rendering pages to images and iterating for layout fidelity; the author’s test produced a polished PDF after self-correction. Codex CLI now loads user skills from ~/.codex/skills, and successfully used one to scaffold a working Datasette plugin—evidence that skills are practical and worth standardizing.

Key Points

OpenAI has introduced filesystem “skills” in ChatGPT (Code Interpreter) and Codex CLI, mirroring Anthropic’s simple folder-with-Markdown pattern.
ChatGPT’s skills include documents and PDFs, using page rendering to PNGs plus vision models to preserve layout and visuals over raw text extraction.
A real-world test produced a high-quality PDF after iterative self-checks, including automatic font substitution to handle macrons in “kākāpō.”
Codex CLI can load user-installed skills from ~/.codex/skills; Willison used this to have Codex generate a functioning Datasette plugin.
The author calls for a formal, minimal skills spec—potentially overseen by the Agentic AI Foundation—because the pattern is gaining rapid adoption.

Sentiment

The overall sentiment of the Hacker News discussion regarding the article is largely positive and pragmatic concerning the 'skills' mechanism itself, recognizing its utility and clever design. However, it quickly branches into a more mixed and often skeptical or philosophical debate when discussing the broader implications for AGI and the nature of AI 'understanding.'

In Agreement

The 'skills' concept is a simple, 'sticky,' and intuitive product innovation, likened to 'the RSS of AI' due to its lightweight and effective design.
Skills are highly practical for concrete use cases like enforcing coding best practices (debugging, commit messages, PRs), managing specialized tasks, and reducing context window pollution for LLMs.
The ability of skills to enable 'computability' (running scripts, accessing external tools/data) is a 'superpower' that compensates for LLM weaknesses and allows for autonomous task execution and self-improvement.
LLMs, such as Claude Opus 4.5, are capable of generating skills themselves, further enhancing their utility and reducing manual effort.
The call for a formal, vendor-neutral specification for skills (e.g., by the Agentic AI Foundation) is supported to ensure cross-vendor compatibility and broader adoption.
The 'quiet' adoption by OpenAI, as highlighted by the article, is seen as noteworthy given the significance of the feature and OpenAI's typical communication style.

Opposed

Skills are essentially an 'automated way to introduce user and system prompt stuffing' into context, implying a level of complexity rather than fundamental innovation.
The use of 'markdown-with-English' as a DSL for skills is criticized as being imprecise, uncheckable, and inefficient compared to formal programming languages, suggesting it moves AI 'further from AGI.'
Current LLMs are not truly 'intelligent' or AGI; they are sophisticated statistical models, and advancements like 'skills' are mere 'quality of life' features or 'library functions' rather than demonstrations of real understanding.
AGI is an 'illusion' or 'mirage,' and the focus on achieving it distracts from delivering practical, domain-specific value.
Benchmarks for AGI are problematic due to potential overfitting or 'hacking,' making real-world application a more reliable metric.
The claim that LLMs 'understand' is merely a semantic debate, as their output is based on statistical models, not genuine comprehension.