---

name: storage-secrets-ingress description: "Use LeptonAI storage/file, secret, ingress routing, access controls, and mount/env/secret parsing safely." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

storage-secrets-ingress

Use this sub-skill when a LeptonAI task involves workspace file storage, secrets, ingress/domain routing, canary endpoint weights, endpoint access-control flags, or validating storage mount/env/secret strings before a workload command.

Route here for

• lep storage ... file-system commands and the hidden lep file ... alias.
• Storage operations: upload, download, rm, rmdir, mkdir, du, ls, and ls-file-system.
• lep secret create/list/remove planning, redaction, and secret-to-env references.
• lep ingress list/create/get/delete/add-endpoint/remove-endpoint/update-endpoint/set-endpoints planning.
• Canary routing with relative ingress weights and safe rollout/rollback command sequences.
• Endpoint access-control review involving --public, --ip-whitelist, and --tokens.
• Mount strings of the form FROM_PATH:MOUNT_PATH:VOLUME, including node-local and node-nfs:<storage_name>.
• Preflight parsing of --mount, --env, --secret, IP allowlist, token, and complete ingress endpoint-list plans without contacting Lepton.

Route elsewhere

• Creating, updating, stopping, deleting, or scaling endpoints/jobs/pods/Ray/fine-tune workloads themselves: use workload-management and return here only for storage, secret, mount, ingress, or access-control details.
• Login, workspace selection, credential persistence, or token refresh: use workspace-and-auth.
• Python SDK endpoint calls and APIClient programming patterns not specific to storage/secret/ingress resource methods: use sdk-client.
• General lep command discovery, shell safety, and top-level CLI behavior: use cli-operations.

Required safety posture

Lepton storage, secret, ingress, and endpoint access-control commands are workspace-scoped cloud operations. Do not run list/create/update/delete/upload/download commands unless the user has authorized the live workspace operation. Before any mutation or data transfer, produce a read-first plan, name the target resource/path/domain, ask for confirmation, and redact tokens and secret values.

Prefer these bundled references:

• storage-and-secrets.md for file-storage and secret workflows.
• ingress-and-routing.md for ingress lifecycle, canary weights, and access-control interplay.
• mounts-env-secrets.md for mount, env, secret, IP allowlist, token, and endpoint-list preflights.
• troubleshooting.md for destructive replacements, path-confirmation errors, redaction, and delete safeguards.

Quick safe preflight

Use the bundled parser before building workload or ingress commands:

python scripts/validate_mounts_env.py \
  --mount ./data:/mnt/data:node-local \
  --mount storage-cache:/cache:node-nfs:my-nfs \
  --env MODE=production \
  --secret API_KEY \
  --ip-whitelist 203.0.113.0/24 \
  --existing-endpoint stable:100 \
  --set-endpoint stable:90 \
  --set-endpoint canary:10

The script performs only local parsing/linting. It never opens a Lepton workspace client and never transfers data or mutates cloud state.

Minimal mutation checklist

1. Identify the exact live target: storage path/file-system, secret name, ingress domain/name, endpoint name, or complete endpoint list.
2. Run a read command first when authorized: lep storage ls ..., lep secret list, lep ingress get -n ..., or the workload status route.
3. Validate local syntax with scripts/validate_mounts_env.py when mounts/env/secrets/IP allowlists or set-endpoints are involved.
4. Ask for explicit confirmation for upload, download overwrites, rm, rmdir, secret remove, ingress delete, remove-endpoint, and especially set-endpoints.
5. Execute only the confirmed command; do not print secret values or literal access tokens in logs.