Claude Skills: Simple Files, Big Agent Power

Claude Skills package task know-how as Markdown plus optional scripts that Claude loads only when needed, making them token-efficient and easy to share. They depend on a sandboxed coding environment, turning Claude Code into a general-purpose agent for real computer automation. Compared to MCP, Skills are simpler, lighter, model-agnostic, and likely to drive a rapid proliferation of community-built capabilities.

Key Points

• Skills are lightweight: a Markdown file with YAML metadata and optional scripts that Claude loads only when relevant, keeping token costs low.
• They require a coding environment (filesystem + command execution), which unlocks powerful automation but demands strong sandboxing and security.
• Claude Code functions as a general-purpose agent for computer automation; Skills make this explicit and extensible for real workflows.
• Compared to MCP, Skills avoid protocol complexity and token bloat—LLMs can learn tools at runtime and use simple, shareable instructions.
• Skills are portable and model-agnostic, enabling broad community sharing and a likely surge in practical, domain-specific capabilities.

Sentiment

Overall, the sentiment of the Hacker News discussion is mixed and largely skeptical/nuanced regarding the 'newness' and absolute superiority of Claude Skills over MCP. While token efficiency and formalization are acknowledged as valuable, many commenters perceive Skills as an iteration on existing concepts or as not fully replacing the distinct functionalities of MCP.

In Agreement

• Skills are effectively 'Just In Time' context injections, leading to token efficiency by only loading relevant content when needed, which is a significant improvement over large, pre-loaded MCPs.
• The formalization and standardization offered by Skills induce efficiency and are a positive evolution, even if the underlying concept isn't entirely new.
• Specific examples, like the Excel+Python skill, demonstrate real value by effectively handling complex edge cases in practical scenarios.
• Using command-line tools and scripts, as Skills enable, can be faster to write and debug compared to some MCP implementations.
• The token bloat of many existing MCPs (e.g., GitHub's 30K tokens) validates the need for a more efficient context loading mechanism like Skills.

Opposed

• Skills are not a significant breakthrough but rather a formalization or 'repackaging' of existing practices like manually linked documentation ('Table of Contents' approach) or Retrieval-Augmented Generation (RAG).
• Existing custom MCP workflows, especially those involving indexed documentation, already cover similar ground and may offer comparable or superior value.
• Skills do not replace the core functionality of MCP, which provides LLMs with access to external APIs; Skills primarily offer contextual information and local script execution.
• MCPs can also achieve token efficiency for 'help' functions or with proper implementation (e.g., using `tools/listChanged`), suggesting the problem is often with how MCPs are used, not the protocol itself.
• Concerns exist about the non-deterministic parsing of Markdown for LLM efficiency, with some arguing that structured schemas like JSON would be more reliable for programmatic consumption.
• Skepticism regarding how an LLM would choose between an MCP tool and a skill, and whether current investments in MCP servers will become obsolete too quickly.