SKILL.md
PulseAugur coverage of SKILL.md — every cluster mentioning SKILL.md across labs, papers, and developer communities, ranked by signal.
16 day(s) with sentiment data
-
AI Agent Skills Pose Security Risks, Audit Recommended
The article warns that agent skills, which are code snippets used by AI agents, should be treated as untrusted code. It suggests a three-step audit process before installing any such skills to mitigate security risks. T…
-
Claude Code: Top 6 Skills for Developers Recommended
A guide recommends six specific Claude Code skills for developers, emphasizing the importance of installing only one skill per development phase to avoid unpredictable behavior. The selection criteria prioritize officia…
-
Anthropic's Claude limits skill context to 500 lines
Anthropic's Claude AI model has a context window limitation that requires users to keep their "Claude Code SKILL.md" files under 500 lines. Bloated skills consume context for the entire session, impacting performance. D…
-
Claude Code users can disable model invocation to save tokens
A reminder for users of Claude Code is to avoid polluting their default session context. By setting `disable-model-invocation: true` within the `SKILL.md` file, users can prevent a specific skill from being included in …
-
New framework clarifies AI agent architecture into five layers
An AI agent's architecture can be understood through five distinct layers: Prompt, Context, Loop, Graph, and Harness. This framework helps developers diagnose and fix issues by identifying which layer is responsible for…
-
Build a regression fixture to test agent skill security
This article proposes a method to test the security of third-party agent skills before integrating them into a system. The author suggests creating a regression fixture that executes a skill within a controlled environm…
-
Anthropic's Claude Code plugin enforces unique website design
Anthropic has released a Frontend Design plugin for Claude Code, designed to prevent the AI from generating generic website aesthetics. This plugin enforces a deliberate design process, requiring Claude to first define …
-
Claude Code upgraded to three-layer agent system with enhanced permissions
The developer behind the Claude Code agent-skills repository has updated it to a three-layer system, renaming it to nasrulhazim/claude. This upgrade addresses limitations encountered with the previous skill-based approa…
-
Claude Code plugin development reveals naming, execution, and filesystem issues
A developer encountered three issues when packaging their Claude Code plugin, two of which stemmed from the plugin's name becoming a prefix for skill calls and the CLI's execution context changing. The fixes involved ex…
-
Genkit Go uses Agent Skills to reduce token costs and context bloat
Genkit Go has introduced Agent Skills, a modular approach designed to reduce context window bloat and token costs. This feature utilizes SKILL.md bundles to separate specialized functionalities from the main context, al…
-
Qwen Code introduces '/learn' feature for saving AI-generated instructions
Qwen Code has introduced a new '/learn' feature that allows users to save specific instructions, rules, or processes derived from conversations or external sources into a separate SKILL.md file. This aims to preserve va…
-
AI skills and plugins need 'motors' to act, not just instructions
The author argues that AI "skills" and "plugins" are often conflated with actual tools, leading to inefficiencies and "context debt" in AI agent workflows. A skill should guide an agent on *how* to use a capability, whi…
-
SKILL.md: Markdown files emerge as powerful, context-aware AI tools
A SKILL.md file, which is essentially a markdown document, is being highlighted as a powerful and underrated tool for enhancing AI capabilities, particularly with models like Claude. This format allows users to package …
-
Claude Code: Optimize AI agent instructions by moving rules out of CLAUDE.md
The author proposes a more efficient method for managing instructions for AI agents like Claude, suggesting that the CLAUDE.md file should be reserved for invariant rules that apply to every interaction. Other types of …
-
Matt Pocock, Vercel Labs Dominate Claude Code Skills Leaderboard
A recent analysis of Claude Code skills reveals that the top 10 most-installed skills are dominated by just three authors: Matt Pocock, Vercel Labs, and Anthropic. Matt Pocock leads with five skills, Vercel Labs has fou…
-
Claude Code's extensibility detailed via Skills, Plugins, and MCP Servers
The articles discuss Claude Code's extensibility through various mechanisms, primarily focusing on "Skills." Skills are described as reusable instruction packs, often in Markdown format (SKILL.md), that extend Claude Co…
-
Reelier enables deterministic AI agent workflows with zero-token replays
Reelier is a new tool designed to address the inefficiency and non-determinism in AI agent workflows. It allows users to record a single successful execution of an agent's task and then replay it deterministically using…
-
Anthropic's Claude Code skills deviate from documentation guidelines
Anthropic's Claude Code documentation suggests skills should be under 500 lines, but the actual implementation of their 17 official skills often exceeds this limit, with some reaching nearly 600 lines. The effectiveness…
-
Activeloop's Hivemind codifies coding patterns for AI agents
Activeloop has released Hivemind, a continual-learning layer for coding agents designed to capture and reuse team members' coding patterns. Hivemind mines repeated actions from agent sessions and codifies them into reus…
-
SKILL.md standard unifies AI coding agent instructions
A new open standard called SKILL.md has been introduced to consolidate instruction files for various AI coding agents, including Claude Code, Codex, and Cursor. Previously, developers had to maintain separate instructio…