From deployshield
Cloud infrastructure and production safety guardrails - prevents write/mutating operations on cloud, database, and deployment CLIs
How this skill is triggered — by the user, by Claude, or both
Slash command
/deployshield:deployshieldThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
DeployShield is active and guarding against accidental write/mutating operations on production environments.
DeployShield is active and guarding against accidental write/mutating operations on production environments.
The following CLIs are monitored. Only read-only commands are permitted:
| CLI | Allowed Actions |
|---|---|
aws | describe, get, list, wait, help, sts get-*, s3 ls, s3 cp (download only), configure list, sso login |
gcloud | describe, list, info, help, config list, config get-value, auth list, auth print-* |
az | show, list, get, help, account show, account list, ad signed-in-user show |
kubectl | get, describe, logs, explain, api-resources, api-versions, cluster-info, version, config view, config get-*, config current-context, auth can-i, diff, top, wait, any command with --dry-run |
helm | list, ls, get, show, status, history, search, repo list, template, lint, verify, version, env, any command with --dry-run |
| CLI | Allowed Actions |
|---|---|
psql | -l (list databases), -c with SELECT/SHOW/DESCRIBE/EXPLAIN/\d commands |
mysql | -e with SELECT/SHOW/DESCRIBE/EXPLAIN queries |
mongosh | --eval with .find(), .count(), .stats(), .explain(), show commands |
redis-cli | GET, KEYS, SCAN, INFO, PING, TTL, HGETALL, ZRANGE, and other read commands |
| CLI | Allowed Actions |
|---|---|
terraform | plan, show, output, state list, state show, state pull, validate, fmt, providers, version, graph, workspace list, workspace show, init |
pulumi | preview, stack ls, stack select, config get, whoami, version, about |
cdk | diff, synth, list, doctor, version |
sam | validate, build, local, logs, list |
serverless/sls | info, print, package, invoke local |
ansible-playbook | Only with --check, --syntax-check, --list-hosts, --list-tasks |
| CLI | Allowed Actions |
|---|---|
vault | read, list, status, version, login, token lookup, kv get |
gh | pr view/list, issue view/list, api (GET), repo view/list/clone, auth, search |
docker/podman | ps, images, logs, inspect, info, version, stats, compose ps/logs |
npm/yarn/pnpm | Everything except publish/unpublish |
twine | Everything except upload |
gem | Everything except push/yank |
cargo | Everything except publish |
By default, DeployShield blocks ALL write operations. To block only in specific contexts (e.g. production), create a .deployshield.json config file:
Config file locations (first found wins):
$DEPLOYSHIELD_CONFIG env var.deployshield.json in CWD~/.deployshield.jsonExample — only block kubectl/helm in prod contexts, and AWS in the production profile:
{
"kubectl": ["prod", "production", "prod-*"],
"helm": ["prod", "production"],
"aws": ["production"],
"terraform": ["production", "default"]
}
Rules:
[] → never blocked (disabled)Context detection: --context/--kube-context flags and kubeconfig for kubectl/helm, --profile/AWS_PROFILE for aws, TF_WORKSPACE/.terraform/environment for terraform, --project/CLOUDSDK_CORE_PROJECT for gcloud, --subscription/AZURE_SUBSCRIPTION_ID for az, --stack/-s for pulumi.
DeployShield provides deep protection through recursive validation:
$(...) and `...`.<(...) and >(...).sudo and env to check the underlying command.bash -c and sh -c.This ensures that bypasses like sudo terraform apply or bash -c "kubectl delete pods --all" are correctly intercepted.
--dry-run flags where applicable (e.g., kubectl apply --dry-run=client)terraform plan instead of terraform apply to preview changespulumi preview instead of pulumi up to preview changesansible-playbook --check instead of running playbooks directlycdk diff instead of cdk deploy to preview changesSELECT queries to inspect database state, never run DDL/DML directlynpx claudepluginhub matanryngler/deployshield --plugin deployshieldPerforms security reviews for risky code changes like sensitive file edits, shell commands, dependencies, CI/CD, and secrets using Clawdstrike MCP tools for policy checks.