GitHub releases agent skills guidance amid MCP security scanner comparisons

On November 13, 2026, at 06:54, the AI & ML section of the GitHub Blog published "Build once, run on any agent: A practical guide to agent skills," introducing a practical guide to agent skills. The guide explains how to package repeatable engineering work into a SKILL.md file so different agents can reuse it. A skill only describes task steps; tools, project context, and permissions remain specific to the host environment that runs it. Instructions written for personal use can later become part of a team's pipeline, keeping task knowledge out of scattered chats or single points of failure.

Two minutes later, at 06:56, the AI section of DEV Community republished a fact comparison of five open-source MCP and agent skill scanners that originally appeared on Agent Security Index. The canonical, updated version, "MCP security scanners compared: 5 open-source tools," was last updated on November 2, 2026. The article advises selecting a scanner across four dimensions, including scope of checks, whether it launches the server under inspection, and where data goes. Facts are drawn from each project's own README and docs, and star counts and release dates were read from GitHub on November 2, 2026. It notes that no lab results exist yet, making this a fact comparison rather than a ranking by test score.

Together, the two posts move agent skills from a published authoring guide to a practice that dedicated security scanners already inspect. Both remain at the level of documentation and factual summaries, with no test-based rankings published so far.