aiops

A curated, directly installable library of SRE and DevOps-infrastructure skills for any AI agent — Claude Code, Codex, Cursor, OpenCode, Kilo, OpenClaw, Hermes, pi, or a shared ~/.agents/skills root.

Catalog

295 validated SKILL.md files: 291 in four default groups, plus 4 in the optional skills/optional/openclaw/ group (4 skill directories, excluded from default installs).

Group	Skills	Coverage
skills/agent-platform/	86	Agent collaboration, MCP, agent evaluation, and agent operations
skills/operations/	148	Operator skill tree — infrastructure/ (Kubernetes, OpenShift including LLM/GPU model serving, SRE, observability), devops/ (Docker/containers, CI/CD including Argo CD, deployment patterns, cloud DevOps), and misc/ (shell tooling)
skills/integrations/	13	Slack, Outlook, GitHub, data-source queries, and cross-source analysis
skills/workflows/	44	Software development, productivity, creative work, social media, smart home, and gaming
skills/optional/openclaw/	4	OpenClaw/NemoClaw-specific skills (skill evaluation, skill-install pipeline, auto-heal) across 4 directories; opt in with --group optional

Each operations/ category root ships a content-rich SKILL.md mapping the category's subskills — purpose, triggers, key commands, and routing guidance — so an agent can survey a whole category from one file before loading a specific subskill. CATALOG.md indexes every skill; skills/CATEGORIES.md is the browsing guide.

Layout

• skills/ — the reviewed skill directories. Keep this layout when copying into an agent skill root.
  • skills/optional/ — agent-specific skills, excluded from default installs.
  • skills/licenses/ — collection-level license files; per-skill LICENSE files stay inside their skill directories.
• CATALOG.md — organized index of every SKILL.md in this repository.
• SOURCE_NOTICES.md — source and license record for redistributed material.
• PENDING_SOURCES.md — requested sources that were not available on the build host.
• SECURITY.md — trust model for the library and vulnerability reporting.
• AGENTS.md — instructions for AI agents working in this repository.
• scripts/validate-skills.sh — metadata validation gate.
• bin/install.js + package.json — the npx installer (plain Node.js, no dependencies).
• .skillspector-baseline.yaml — reviewed security-scan findings with written justifications.

Install

Install with npx (Node.js >= 16, zero dependencies), selecting your agent:

Agent	Default skill directory	Install
Claude Code	~/.claude/skills	npx github:anuguda/aiops --agent claude
Codex	~/.codex/skills	npx github:anuguda/aiops --agent codex
Cursor	~/.cursor/skills	npx github:anuguda/aiops --agent cursor
OpenCode	~/.config/opencode/skills	npx github:anuguda/aiops --agent opencode
Kilo	~/.kilo/skills	npx github:anuguda/aiops --agent kilo
OpenClaw	~/.openclaw/skills	npx github:anuguda/aiops --agent openclaw
Hermes	~/.hermes/skills	npx github:anuguda/aiops --agent hermes
pi (omp)	~/.omp/agent/skills	npx github:anuguda/aiops --agent pi
shared root	~/.agents/skills	npx github:anuguda/aiops --agent agents

The target is the skill root — the directory whose direct children are the installed groups. Resolution order: --target <dir>, --agent <name>, $AIOPS_SKILLS_DIR, legacy $HERMES_SKILLS_DIR / $HERMES_HOME/skills, or an interactive prompt when a TTY is available. Existing files are skipped (pass --force to overwrite) and every copied SKILL.md is validated, matching scripts/validate-skills.sh. An unscoped install excludes skills/optional/ and skills/licenses/.

Inspect and install selectively:

npx github:anuguda/aiops --list                         # groups + skill counts
npx github:anuguda/aiops --group operations            # one group (repeatable)
npx github:anuguda/aiops --group optional              # OpenClaw/NemoClaw skills
npx github:anuguda/aiops --skill openshift-llm-deploy  # only the named skill(s)
npx github:anuguda/aiops --group operations --skill slack-channel-finder

Selections compose: --skill always searches all groups, so it adds cross-group skills (including optional ones) on top of a --group selection. Named skills install under their full group/category path (operations/infrastructure/openshift/openshift-llm-deploy), so they overlay cleanly on top of or under a later group install. --dry-run previews without writing.

To install manually, copy the required directories from skills/ into your agent's skill root, then start a new agent session. Groups are copied whole, so per-skill LICENSE files travel with their skills; a full (unscoped) install also copies the collection-level license files from skills/licenses/ to licenses/ plus the repository's Apache-2.0 LICENSE to the target root. Repository and Finder metadata (.gitignore, .npmignore, .DS_Store) is never copied, so local installs match npm-packaged installs file-for-file. The package installs from the git URL; once published to the npm registry the same command shortens to npx aiops-skills.

Validate a tree before publishing or installation:

scripts/validate-skills.sh [skills-root]

The script exits non-zero if any SKILL.md is missing its Markdown title.

Agents discover skill commands recursively, and all deployment gateway authorization rules continue to apply. Deployment-specific credentials and policy configuration belong outside this repository.

Security

Every skill in this library is scanned with SkillSpector, an agent-skill vulnerability scanner. CI (.github/workflows/skillspector.yml) runs the scan on each pull request and fails on any new finding.

Run it locally before submitting a change:

pip install "git+https://github.com/NVIDIA/skillspector.git@v2.11.1"
skillspector scan . --no-llm --baseline .skillspector-baseline.yaml

• SECURITY.md states the trust model these skills are held to and how to report a vulnerability.
• .skillspector-baseline.yaml records reviewed false positives, each with a written justification. Findings in executable files are triaged individually rather than suppressed by pattern.

Skills carry no credentials. Each integration names the environment variables it needs (for example GITHUB_TOKEN, SLACK_BOT_TOKEN) and reads only those keys; the operator supplies the values outside this repository.

Licensing

This repository is licensed under Apache-2.0; first-party skills are authored by Anurag Guda. Every active skill is MIT- or Apache-2.0-licensed. Skills derived from upstream projects keep their original licenses, recorded in SOURCE_NOTICES.md with the collection-level license files under skills/licenses/; non-redistributable or environment-bound material was removed or replaced with clean-room first-party counterparts.

Provenance

This repository preserves the source notices and license files shipped with the skills it redistributes. It is an organized distribution of the aiops skill library, not a replacement for upstream projects.