ops-engineering-skills

What it claims

A community-built (not Anthropic-affiliated) library of 296 Agent Skills across 22 independently-installable domain plugins. Beyond breadth, the page establishes two design rules the summary doesn't: (1) generic concept skills (e.g. sast-integration) are kept separate from tool-specific skills (e.g. sonarqube-code-quality-and-security, fortify-static-analysis), which cross-reference the concept skill rather than re-explaining it — so an agent gets conceptual altitude when someone is deciding whether to adopt a practice, tool-specific detail when they're configuring one; (2) scope is deliberately confined to ops engineering, explicitly not general software dev, data analytics, or writing.

How to use it

Four install paths, all producing identical content. Claude Code gets a native marketplace with version tracking; the domains are independent, so install only what your stack needs:

/plugin marketplace add selvarajmurugesan90/ops-engineering-skills
/plugin install devsecops-skills@ops-engineering-skills

Update with /plugin marketplace update + /plugin update <plugin>@ops-engineering-skills.

For any compliant agent via the community CLI:

npx skills add selvarajmurugesan90/ops-engineering-skills --list
npx skills add selvarajmurugesan90/ops-engineering-skills \
  --skill pod-crashloop-and-oom-troubleshooting --agent claude-code -y
npx skills add selvarajmurugesan90/ops-engineering-skills --all

For Codex/Copilot/Cursor/Gemini via shell script (--plugin takes a domain folder name or all):

./installers/install.sh --plugin all --target .codex/skills
./installers/install.sh --plugin devsecops --target .github/skills

Or plain cp -r of any folder containing a SKILL.md. Skills activate on plain-language request matching against each skill's description frontmatter — no invoke syntax. Licence is Apache-2.0, commercial use permitted with attribution.

Evidence of quality

The validator (scripts/validate_skills.py) enforces per-change: correct frontmatter, kebab-case names ≤64 chars, metadata.domain matching folder, all 8 sections present and ordered, duplicate-name detection across all 296, and a hardcoded-secret scan. Cross-reference links are checked to resolve (page claims zero broken links). A destructive-action review flags every irreversible command (terraform destroy, kubectl delete namespace, force-push, etcd restore, TRUNCATE/DROP KEYSPACE) with a warning and safer alternative rather than as a default.

Limits

Cross-agent support is verified only via claude plugin validate and installer file-placement correctness — not live install-and-invoke sessions in Copilot, Codex, Cursor, or Gemini CLI (listed as future work). Skill currency is self-reported through a maturity field (stable/draft) and version ranges cited in each skill's Prerequisites section rather than pinned versions; you must check these yourself. One skill, code-knowledge-graph-tools-for-ai-agents, describes GitNexus, whose free tier carries a non-commercial licence (applies to that tool, not the repo).