Aws vault
Retrieve application secrets from AWS Secrets Manager and run strictly READ-ONLY kubectl commands via aws-vault. Handles environment-specific profiles and secret naming conventions resolved through a user-maintained service registry. TRIGGER when: fetching API keys, credentials, or any secret stored in AWS Secrets Manager; running read-only kubectl operations (get, describe, logs, top, events, etc.) against an aws-vault-protected cluster. DO NOT TRIGGER when: secrets are available in the local environment, the target is not an AWS-backed service, or you need to MUTATE cluster state (apply, create, delete, scale, exec, port-forward, etc.) — those run only under direct supervisor control in a separate terminal.From its SKILL.md
npx -y skills add Dev10x-Guru/Dev10x-Claude --skill aws-vaultAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 8 stars8 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
9.7 KB, as published. Nobody here has run it
AWS Vault Secret Retrieval
When to Use
- Retrieving API keys or credentials for SaaS integrations
- Comparing secrets across staging and production environments
- Discovering available secrets for a service
- Running read-only kubectl operations against a cluster gated
by aws-vault credentials (
get,describe,logs,top,events,explain,version,cluster-info,api-resources,api-versions,auth,wait,diff)
Critical Rules
Secrets are NOT in the local environment. SaaS API keys are stored
in AWS Secrets Manager and injected into k8s pods at runtime. Never
use echo $VAR or printenv to look for them. Always use the wrapper
scripts.
REQUIRED: Call AskUserQuestion before accessing any secret (do
NOT use plain text). Prompt: "I need to access the <secret-id>
secret in <env> — is that OK?" Only proceed after the user
approves. This applies even to read-only discovery lookups.
Never call aws secretsmanager directly. Use
${CLAUDE_PLUGIN_ROOT}/skills/aws-vault/scripts/secrets.sh instead.
Never call kubectl or aws-vault exec ... kubectl directly.
Use ${CLAUDE_PLUGIN_ROOT}/skills/aws-vault/scripts/kubectl.sh
instead.
Never call aws or aws-vault exec ... aws directly. Use
${CLAUDE_PLUGIN_ROOT}/skills/aws-vault/scripts/aws.sh instead. It
is strictly read-only: only describe-*, list-*, get-*,
lookup-*, search-*, head-*, batch-get-*, view-*, and
estimate-* operations pass through. Secret-exfil reads
(secretsmanager get-secret-value, ssm get-parameter --with-decryption, kms decrypt, sts get-session-token,
ec2 get-password-data, ecr get-login-password, …) are denied
even though their verb is "get" — fetch secrets through
secrets.sh under the AskUserQuestion gate instead.
The kubectl wrapper is strictly read-only. Only the verbs
listed under "When to Use" pass through; mutating verbs
(apply, create, delete, patch, scale, rollout,
exec, port-forward, proxy, cp, debug, set,
label, annotate, taint, drain, cordon, run,
expose, autoscale, ...) are denied. Privilege-escalation
and cluster-redirection flags (--as, --as-group,
--token, --server, --kubeconfig,
--insecure-skip-tls-verify) are denied anywhere in the
argument list so the verb allowlist cannot be bypassed via
flag injection.
When the agent attempts a denied verb, the wrapper exits non-zero and prints a copy-pasteable snippet (with profile, context, namespace, and verb already substituted) for the supervisor to run in a separate terminal under direct control. The agent MUST NOT carry that snippet back through any other tool — surface it to the user and stop.
Service Registry
The scripts resolve environment/service mappings from a user-maintained YAML registry at:
~/.config/Dev10x/aws-vault/service-registry.yaml
The registry maps:
- Environment name →
aws_vault_profile, k8scontext,namespace - Service name → secret ID under
environments.<env>.secrets.<service> - Common key names →
secret_keys.<key_name>(informational)
If the file is missing, copy the example from
${CLAUDE_PLUGIN_ROOT}/skills/aws-vault/references/service-registry.example.yaml
to the path above and edit it for your environments.
Workflow: Retrieve a Secret
Step 0: Ask user confirmation
Call AskUserQuestion with the secret ID and environment. Only
proceed after explicit approval.
Step 1: Resolve the registry entry
Read ~/.config/Dev10x/aws-vault/service-registry.yaml to confirm
the environment and service key. If a mapping is missing, run the
Discover Secrets workflow first, then update the registry.
Step 2: Retrieve the secret via wrapper
${CLAUDE_PLUGIN_ROOT}/skills/aws-vault/scripts/secrets.sh <env> <service> --key <KEY_NAME>
Examples (replace <env> and <service> with values from your
registry):
${CLAUDE_PLUGIN_ROOT}/skills/aws-vault/scripts/secrets.sh staging api --key DATABASE_URL
${CLAUDE_PLUGIN_ROOT}/skills/aws-vault/scripts/secrets.sh production worker
Step 3: Use the secret
Pass the value to downstream commands (curl, SDK calls, etc.). Never log or echo the raw secret value.
Workflow: Discover Secrets
When the secret ID or key name is unknown:
-
Confirm via
AskUserQuestion. -
Run without
--keyto retrieve the full JSON blob:${CLAUDE_PLUGIN_ROOT}/skills/aws-vault/scripts/secrets.sh staging api -
Pipe through
jq 'keys'(in a separate command) to list available keys without exposing values. -
Update the registry at
~/.config/Dev10x/aws-vault/service-registry.yamlundersecret_keysso the mapping is reusable.
Workflow: kubectl Operations (read-only)
${CLAUDE_PLUGIN_ROOT}/skills/aws-vault/scripts/kubectl.sh <env> <verb> [args...]
The wrapper resolves context and namespace from the registry per
environment, then enforces a verb allowlist + flag deny-list before
invoking aws-vault exec ... -- kubectl ....
Examples (all read-only):
${CLAUDE_PLUGIN_ROOT}/skills/aws-vault/scripts/kubectl.sh staging get pods
${CLAUDE_PLUGIN_ROOT}/skills/aws-vault/scripts/kubectl.sh staging logs deploy/<svc> --tail=100
${CLAUDE_PLUGIN_ROOT}/skills/aws-vault/scripts/kubectl.sh production describe pod/<name>
${CLAUDE_PLUGIN_ROOT}/skills/aws-vault/scripts/kubectl.sh staging top pod
${CLAUDE_PLUGIN_ROOT}/skills/aws-vault/scripts/kubectl.sh production auth can-i get pods
What is allowed
| Verb | Use |
|---|---|
get | List/read resources, optionally with selectors |
describe | Verbose read; reveals Secret/ConfigMap names and structure |
logs | Container logs (incl. --follow, --previous) |
top | Pod/node metrics |
events | Cluster event stream |
explain | Schema reference |
version | Client/server version |
cluster-info | Cluster endpoints |
api-resources, api-versions | API discovery |
auth | auth can-i and read-only auth probes |
wait | Block until a resource condition is met |
diff | Compare a local manifest to live state (no mutation) |
What is denied
Mutating verbs (apply, create, delete, patch, scale,
rollout, exec, port-forward, proxy, cp, debug,
set, label, annotate, taint, drain, cordon, run,
expose, autoscale, ...) exit non-zero with a snippet for
manual execution.
Privilege-escalation and cluster-redirection flags are denied anywhere in the argument list — even with an allowed verb:
| Flag prefix | Why denied |
|---|---|
--as, --as-group | RBAC impersonation |
--token | Alternate credentials, bypasses aws-vault |
--server | Points at a different cluster |
--kubeconfig | Loads arbitrary kubeconfig |
--insecure-skip-tls-verify | Disables cert validation |
Caveat on sensitive reads
describe secret and get secret -o yaml are technically
read-only from the API's perspective but reveal base64-encoded
secret values. Treat them as you would a secrets.sh call:
the same AskUserQuestion confirmation gate applies before
invocation.
Workflow: AWS CLI Operations (read-only)
${CLAUDE_PLUGIN_ROOT}/skills/aws-vault/scripts/aws.sh <env> <service> <operation> [args...]
The wrapper resolves the aws_vault_profile from the registry per
environment, enforces a read-operation allowlist + secret-exfil
denylist, then invokes aws-vault exec ... -- aws ....
Examples (all read-only):
${CLAUDE_PLUGIN_ROOT}/skills/aws-vault/scripts/aws.sh staging ec2 describe-instances
${CLAUDE_PLUGIN_ROOT}/skills/aws-vault/scripts/aws.sh production s3api list-buckets
${CLAUDE_PLUGIN_ROOT}/skills/aws-vault/scripts/aws.sh staging logs describe-log-groups
${CLAUDE_PLUGIN_ROOT}/skills/aws-vault/scripts/aws.sh production sts get-caller-identity
Denied (exit non-zero with a copy-pasteable snippet): any mutating
operation (create-*, delete-*, put-*, run-*, terminate-*,
…) and the secret-exfil reads listed under Critical Rules. Run those
in a separate terminal under your own supervision.
Key Lessons
Secret IDs vary by environment
Cluster names and secret IDs differ between environments. Always resolve from the registry rather than hardcoding paths. If the registry entry is missing, run the discovery workflow first, then update the registry.
aws-vault profiles are per-environment
Each environment has its own IAM profile. The profile name is in the
registry under aws_vault_profile.
Timeout on first use
aws-vault exec may prompt for MFA or open a browser for SSO on
first use. Use --debug if authentication fails silently:
aws-vault exec --debug <profile> -- aws sts get-caller-identity