Skip to content

About

No description, website, or topics provided.

Resources

Stars

1 star

Watchers

0 watching

Forks

Latest commit

 

History

17 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 

Repository files navigation

SRE Skills for Claude and other LLM Agents

Claude Skills for Site Reliability Engineers, DevOps, and Cloud engineers focused on monitoring and observability.

Documentation

Skills

Skills are in skills/ (symlinked to .claude/skills/).

Logs

  • gcp-logs
  • grafana-loki
  • nr-logs

Metrics

  • gcp-metrics
  • grafana-mimir
  • nr-metrics

Traces

  • gcp-traces
  • grafana-tempo

Alerts

  • gcp-alerts
  • grafana-alerts
  • nr-alerts

Skill Structure

skills/<skill-name>/
├── SKILL.md     # Instructions with YAML frontmatter (for Claude Code / human use)
├── tool.yaml    # Machine-readable tool definitions (for sre-bot / programmatic use)
└── scripts/
    ├── query.sh # Data fetching (NR skills: + URL + clipboard + FACET tables; --raw for JSON)
    └── url.sh   # UI URL generation

SKILL.md Format

---
name: skill-name
description: What it does. When to use it.
---

# Skill Name

[Concise instructions, under 500 lines]

tool.yaml Format

tool.yaml is the machine-readable counterpart to SKILL.md. While SKILL.md teaches an interactive LLM (like Claude Code) how to use the scripts via natural language, tool.yaml declares the same information as structured data so that automated systems can load and call the scripts without free-form interpretation.

Each tool.yaml defines one or more tools. A tool maps to a single script and declares its name, description, and typed parameters. The parameter order in tool.yaml matches the positional argument order of the underlying script.

tools:
  - name: gcp_logs_query                    # Unique tool name (snake_case)
    description: "Query GCP Cloud Logging"  # When and why to use this tool
    script: scripts/query.sh                # Relative path to the script
    parameters:
      - name: project                       # Maps to $1
        type: string
        required: true
        description: "GCP project ID"
      - name: filter                        # Maps to $2
        type: string
        required: true
        description: "Cloud Logging filter expression"
      - name: limit                         # Maps to $3 (optional, has default)
        type: integer
        default: 1000
        description: "Max entries to return"

Parameter fields:

  • name — Argument name (used by the caller, mapped to positional arg by order)
  • type — string, integer, number, or boolean
  • required — If true, the caller must provide this argument
  • default — Value used when the caller omits the argument
  • description — What this argument does
  • enum — Optional list of allowed values

Conventions:

  • Tool names are prefixed with the skill name in snake_case (e.g., gcp_logs_query, grafana_loki_url)
  • Parameter order must match the script's positional arguments ($1, $2, ...)
  • Optional parameters with defaults should come after required parameters
  • Descriptions should be specific enough for an LLM to choose the right tool and pass correct arguments

Design Principles

  1. Scripts for reliability – Put API calls, auth, URL encoding in executable scripts
  2. Save before analyze – Fetch data to temp files, then analyze (avoids re-fetching)
  3. Compact output – Prefer CSV over JSON when schema is known or when working with many logs (~15x smaller)
  4. Generate UI URLs – Always provide URLs for humans to view in native UIs
  5. Use standard tools – grep, awk, jq, sed, python for analysis

About

No description, website, or topics provided.

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages