Let Claude Code help with infrastructure and on-call without letting it run terraform apply or kubectl delete on its own.
Ten Claude Code subagents for infrastructure and on-call work, plus an infra guard hook that makes Claude ask before terraform apply, kubectl delete, helm upgrade or a cloud CLI change, and refuse terraform destroy -auto-approve, namespace deletes and recursive bucket deletes outright. Every reviewer is read-only by design; the agents that write (CI fixer, runbook and postmortem writers) are limited to their own files.
| Subagent | Use it for | Tools |
|---|---|---|
| incident-triage | live incident: impact, what changed, ranked causes, safest mitigation, status update for the channel | read-only + Bash |
| log-investigator | first error, spikes, error groups, one request traced across services | read-only + Bash |
| terraform-reviewer | plan review before apply: deletes/replacements, IAM wildcards, open security groups, encryption, pinning | read-only + Bash |
| k8s-manifest-reviewer | requests/limits, probes, securityContext, :latest, PDBs, RBAC, exposure (renders Helm/Kustomize first) | read-only + Bash |
| dockerfile-reviewer | base image pinning, non-root, secrets in layers, multi-stage, caching, compose risks | read-only + Bash |
| ci-pipeline-fixer | red CI: root cause from the failing log and the smallest fix; hardens GitHub Actions (SHA pinning, permissions, injection) | edits CI files |
| alert-reviewer | noisy or useless pages, missing for:, runbook links, SLO burn-rate alerts | read-only + Bash |
| runbook-writer | runbooks under docs/runbooks/ with commands that match your repo | writes docs |
| postmortem-writer | blameless postmortem with UTC timeline and action items | writes one doc |
| cloud-cost-reviewer | idle and oversized resources, storage lifecycle, NAT and transfer traps, log retention | read-only + Bash |
Separate from our general Claude Code Subagents Pack (code review, tests, debugging): no agent overlaps.
The screenshot is real output from the kit’s own commands. Tested from a fresh download of this exact zip: ./run_tests.sh ran 10 tests OK on Python 3.13 and 3.10.
Then ask, for example: "Use the terraform-reviewer subagent on the plan for infra/prod", or "@incident-triage checkout is returning 502s since 14:10".
hooks/hooks.json registers hooks/infra_guard.py as a PreToolUse hook on Bash. It reads the command (including cd x && ..., sudo, VAR=1 cmd and bash -c "...") and answers:
Settings: DEVOPS_GUARD=off or DEVOPS_GUARD=deny in the environment, or .claude/devops-guard.json in the project: {"mode": "ask", "allow": ["kubectl apply -f k8s/dev"]} (allow entries are command prefixes). Plugin hooks apply to the whole session while the plugin is enabled. Claude Code ignores hooks, mcpServers and permissionMode in plugin agent files, so the guard ships at plugin level.
./demo.sh shows the linter, the plan summary on examples/plan.json and guard decisions. ./run_tests.sh runs the tests. examples/k8s-deployment.yaml gives the Kubernetes reviewer something to find.
Zip, 31 KB, v1.0, 23 files. You may use and modify this kit for your own work and for clients you serve, including paid client work. Not affiliated with or endorsed by Anthropic, HashiCorp or the Kubernetes project.
10 Claude Code subagents for DevOps and SRE: incident triage, Terraform plan review, Kubernetes, CI fixes, alerts, postmortems. Guard hook blocks destroy. $19.
10 Claude Code subagents for data teams: SQL and dbt review, query tuning, data quality, metrics, A/B tests. SQL guard blocks DROP and unfiltered DELETE. $19.
View Data & Analytics Subagents for Claude CodeMCP server for your own docs: Markdown, MDX, RST, HTML and llms.txt, searched by section with citations. Read-only, offline, no API keys. Claude, Cursor.
View Docs Search MCP ServerTested Claude Code hooks as a plugin: block rm -rf and force-push to main, protect .env and lockfiles, catch API keys in files, audit log. $19 once.
View Claude Code Safety Hooks