Blog
What we're shipping, what we're learning, and how real teams run their work on shared skills.
The skill format is shared across the major coding agents, but the folder each one reads from is not. Codex reads .agents/skills; Claude Code reads only .claude/skills; Cursor and Copilot read both. shareskills now picks the right folder for the tool that is asking, and tells you which one it picked.
Read the post
MCP connectors give an agent access to your CRM, analytics and CMS. They do not give it your team’s way of using them. The three steps from connected to genuinely useful are: stack the connectors, wrap the workflow in a skill, then put it on a schedule. The hard part is doing that as a team.
Read the post
shareskills now runs deterministic validation on every skill version — leaked secrets, personal data, invisible-Unicode smuggling and untriggerable descriptions — with findings surfaced as fixable suggestions, a hard gate on publishing to the directory, and a Checks passed badge on listings.
Read the post
New organisations land in a five-step wizard that explains shareskills-start, asks how your team wants syncing to behave, and seeds the skill into your library with your answers in a config file — while the skill itself stays canonical and keeps receiving updates.
Read the post
Anthropic has shipped peer-to-peer skill sharing inside Claude, and it is good. It is also an inbox, not a library: view-only copies, no review, no history, contents invisible to admins, and only inside Claude. Here is what an organisational intelligence layer still needs, and where the native feature is the right call.
Read the post
Skills can finally be removed from a team library outright: who can delete, how it works from the web and from an agent, and why copies your teammates already installed are deliberately left alone.
Read the post
Skills are becoming the atomic unit of AI-native organisations, and the evidence says they work. The same evidence says ungoverned skill libraries decay exactly the way ungoverned microservices did. Here is the case, the data, and what our own library already shows at 33 skills.
Read the post
Most people in large organisations are still copy-pasting into a chat window. Here is the next gear: agents that work where your files are, and a simple rule for when a task deserves a skill.
Read the post
Cowork can turn a ten-minute screen recording into a reusable skill. Here is how to record one worth keeping — and how to get it off your laptop and into your team’s library.
Read the post
The new cross-vendor plugin standard ships with a hole where distribution and trust should be — on purpose. Here is our first move into it: one-click conformant exports.
Read the post
Malicious AI skills reached 1.7M installs by shipping useful v1s and poisoning later updates. Here is how shareskills now gates exactly that move.
Read the post
The team behind shareskills runs its client work on the product. Here is the actual library — the meeting pipeline, the guardrails skill, and the patterns worth stealing.
Read the post
Written by The shareskills team. New here? Start with how shareskills works.