
Chainguard opened its 1,000+ hardened Agent Skills registry to all users on June 17 — every skill from skills.sh runs through a continuous security pipeline with a per-skill HARDENING.md audit log before it reaches Claude Code, Cursor, Codex, or Gemini CLI.
npx skills add you've run without a second thought.HARDENING.md audit log showing exactly what was changed and why.
chainctl CLI v0.2.282 or later, and a Chainguard Console account (free; no paid plan required). 5# Browse the public catalog
chainctl skills list --group chainguard --recursive
# Pull a skill locally to inspect HARDENING.md before committing
chainctl skills pull chainguard/github/add-educational-comments .
# Install directly — auto-detects Claude Code, Cursor, Codex, Gemini CLI
chainctl skills install chainguard/github/add-educational-commentsinstall command detects every supported agent on your machine and drops the skill into the right directory for each one (e.g. .claude/skills/ for Claude Code). Skills are stored as OCI artifacts at skills.cgr.dev/chainguard. 56HARDENING.md, install:code-review, security-review, find-bugs, postgresql-code-reviewvue-*, react-*, nestjs-*doc-coauthoring (from Anthropic), plantuml-asciideploy-to-vercelgithub/add-educational-comments — adds teaching annotations to code files. The raw skills.sh version carried implicit permissions to execute code. Chainguard's HARDENING.md flagged this as a minimal-permissions violation (severity: high) and stripped the execution permission. The skill only needs to read and annotate files. 5doc-coauthoring (Anthropic) — structured documentation co-authoring. HARDENING.md shows a clean scan — no changes needed, because the upstream source was Anthropic's own well-maintained skill. The audit log still exists; it just records zero issues. 5postgresql-code-review — community skill from skills.sh. Passes through the full hardening pipeline before reaching you. 7skills.cgr.dev/<org>/<name>:<tag>, gets them hardened, and distributes them within the org with SHA pinning for version locking. 8 A "hardening as a service" closed beta is open for teams that want to submit skills built outside skills.sh. 1HARDENING.md is your primary verification tool — read it before installing any skill you plan to use on production code.| Registry | skills.cgr.dev/chainguard |
| Skill count | 1,000+ (from skills.sh) |
| License | Per-skill (mirrors upstream; mostly MIT) |
| Install CLI | chainctl v0.2.282+ |
| Agents supported | Claude Code, Cursor, GitHub Copilot (Codex), Gemini CLI |
| Account required | Chainguard Console (free) |
| Audit trail | HARDENING.md per skill |
| Hardening model | Continuous — reruns on upstream changes |
| Private registry | Available; hardening-as-a-service in closed beta |
| Announced | March 17, 2026 (beta) → June 17, 2026 (public) |
pull-before-install pattern gives you a chance to review every change before it runs in your session. Teams using Claude Code or Cursor on production codebases have the most to gain — a compromised skill with access to your codebase and filesystem is a real attack vector, and Chainguard directly addresses it.chainguard.dev
chainguard.dev
thenewstack.io
edu.chainguard.dev
youtube.com
chainguard.dev
edu.chainguard.dev
x.com
sourceforge.net
github.com
This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.