Expert SRE investigator for incidents and debugging. Uses hypothesis-driven methodology and systematic triage. Can query Axiom observability when available.

What It Does

Hypothesis-Driven Investigation - State, test, disprove hypotheses with data queries
Systematic Triage - Golden signals (traffic, errors, latency, saturation), USE/RED methods
Memory System - Persistent knowledge base for patterns, queries, facts, and incidents
Axiom Integration - Query logs, generate shareable links, discover schemas

Installation

# Amp
amp skill add axiomhq/skills/sre

# npx (Claude Code, Cursor, Codex, and more)
npx skills add axiomhq/skills -s sre

Prerequisites

Access to Axiom deployment(s)
Tools: jq, curl

Setup

Run the interactive setup to configure Axiom access and initialize memory:

scripts/setup

This will:

Create the memory system for storing patterns and learnings
Guide you through creating ~/.axiom.toml if it doesn't exist

To configure manually, create ~/.axiom.toml:

[deployments.prod]
url = "https://api.axiom.co"
token = "xaat-your-api-token"
org_id = "your-org-id"

Get your org_id from Settings → Organization. For the token, create a scoped API token (Settings → API Tokens) with the permissions your workflow needs. Avoid Personal Access Tokens for automated tooling.

Usage

The skill activates for incident response, root cause analysis, production debugging, or log investigation. Key scripts:

# Run APL queries
scripts/axiom-query <deployment> "<apl query>"

# Make API calls
scripts/axiom-api <deployment> GET "/v1/datasets"

# Generate shareable query links
scripts/axiom-link <deployment> "<apl query>" "<time range>"

# Setup personal memory tier
scripts/setup

Scripts

Script	Purpose
`axiom-query`	Run APL queries against Axiom
`axiom-api`	Make raw API calls
`axiom-link`	Generate shareable query URLs
`axiom-deployments`	List configured deployments
`setup`	Initialize memory system
`mem-write`	Write entries to memory KB
`mem-sync`	Sync org memory from git
`mem-digest`	Consolidate journal to KB
`mem-doctor`	Health check all memory tiers
`mem-share`	Push org memory changes

Key Principles

Never guess - query to verify
State facts, not assumptions
Disprove hypotheses, don't confirm
Time filter FIRST in all queries
Discover schema before querying unfamiliar datasets