{
"$type": "site.standard.document",
"bskyPostRef": {
"cid": "bafyreifowyorrl4xclz2fugn3bb6htg74cno4pd3bo6gynkyhecwmm3am4",
"uri": "at://did:plc:2nqvpjte5km46k6dbovaxito/app.bsky.feed.post/3mofad3tmq6l2"
},
"coverImage": {
"$type": "blob",
"ref": {
"$link": "bafkreihqibigz2ua4thv27hvvmznr5nmkuh2qbpazit2libgmpuqosgzxq"
},
"mimeType": "image/webp",
"size": 218238
},
"description": "Three open-source managers for AI agent skills each hit 2,000 GitHub stars in months. But a skill is natural-language intent an agent runs with full file and shell access, and only one of the three scans skill files for attacks before an agent reads them.",
"path": "/ai-agent-skill-managers-are-a-supply-chain-surface-most-of-them-dont-guard/",
"publishedAt": "2026-06-16T07:05:27.000Z",
"site": "https://www.implicator.ai",
"tags": [
"Subscribe for $8/month",
"Subscribe now"
],
"textContent": "\n_Implicator PRO Briefing / Tuesday, June 15 2026_\n\n---\nPro Members Only In five months, three open-source tools turned AI agent skills into a managed package layer, and each crossed 2,000 GitHub stars. The convenience is obvious: edit a skill once, sync it to Claude Code, Codex, Cursor, and a dozen other agents at once. The risk is quieter. A skill is natural-language intent an agent runs with full file, shell, and network access, which makes every one of these managers a supply-chain channel for executable instructions. This briefing compares skillshare, skills-manager, and skills-manage on architecture, maintenance, security, and standard compliance, and shows why only one of them treats the skill file as a threat. New to Implicator PRO? Subscribe for $8/month — new deep dive every Tuesday morning 3am PST.\n\n### This post is for subscribers only\n\nBecome a member to get access to all content\n\nSubscribe now",
"title": "AI Agent Skill Managers Are a Supply-Chain Surface Most of Them Don't Guard",
"updatedAt": "2026-06-16T09:05:37.408Z"
}