Deploy compatible server, static-web, worker, scheduled-job, or reviewed remote-desktop workloads from GitHub or local source to Sealos Cloud, then run the default Runtime Truth Pass against the returned App URL, public route, authentication flow, logs, database state, and full resource footprint. Reject unsupported desktop, mobile, CLI, library, extension, hardware-dependent, mixed, and unidentified targets before readiness scoring or build. Use when the user asks to deploy a repository to Sealos or another cloud platform, or invokes "/sealos-deploy".
يبقى الأمر في سطر واحد. مرّر أفقيًا لمراجعته كاملًا قبل النسخ.
تفضّل نسخة محلية؟ نزّل الملفات المتاحة حاليًا لدى SkillsMP.
مستكشف الملفات
47 ملفات
عرض SKILL.md
SKILL.md
تعليمات المصدر · معاينة للقراءة فقط
name
sealos-deploy
description
Deploy compatible server, static-web, worker, scheduled-job, or reviewed remote-desktop workloads from GitHub or local source to Sealos Cloud, then run the default Runtime Truth Pass against the returned App URL, public route, authentication flow, logs, database state, and full resource footprint. Reject unsupported desktop, mobile, CLI, library, extension, hardware-dependent, mixed, and unidentified targets before readiness scoring or build. Use when the user asks to deploy a repository to Sealos or another cloud platform, or invokes "/sealos-deploy".
metadata
{"author":"labring","compatibility":"Sealos auth/workspace are required for deploys. Docker, buildx, and gh CLI are required only when the selected path needs local build/push. git is required when cloning from a GitHub URL or when git metadata is needed. Node.js 18+ remains an optional accelerator. Phase 5 requires Python 3.8+ with PyYAML; root Compose conversion also requires kompose and may require crane when image tags are floating."}
Sealos Deploy
Identity and Discovery
Owner:sealos-deploy (/sealos-deploy and deploy, update, publish, or cloud-runtime requests).
Class:composite-orchestration across readiness, Dockerfile, template, build, deployment, and Runtime Truth.
Canaries:DEP-KUBECONFIG-SCOPE, DEP-CONFIRM-MUTATION, DEP-REDACT, and DEP-RUNTIME-TRUTH.
Contract: Read references/deploy-contract.md after the canaries pass. It defines the typed phase handoffs, owned artifacts, terminal states, and the read-only Canvas boundary.
.sealos
Scope and Boundaries
Accept a local path or GitHub URL and scope all work to the selected namespace/app. Preserve the current DEPLOY/UPDATE phase order, .sealos artifact inventory, one log file, dependency handoffs, and cleanup footprint. All Kubernetes commands use KUBECONFIG=~/.sealos/kubeconfig kubectl --insecure-skip-tls-verify; unsupported workloads stop before scoring/build.
Risk and Confirmation
Keep auth/workspace, kubeconfig scope, system-tool installation, public exposure, credential changes, deletion, rollback, and cleanup confirmation visible before module detail. Redact passwords, tokens, cookies, env values, kubeconfig, Secret data, and full connection strings. A quality gate and actual Runtime Truth evidence are acceptance conditions.
For every gated mutation, report the exact operation, impact, confirmation, and post-action evidence. Keep sanitized logs, state, diagnostics, and footprint evidence free of secret data.
Lifecycle Workflow
For each request, run preflight/auth/workspace, detect mode, enforce eligibility, assess/detect/build or reuse, generate/validate the template, deploy or update, run Runtime Truth, and record state/cleanup evidence. Emit request-scoped success, stopped, or error; the existing phase modules below remain authoritative for detailed behavior.
Progressive Disclosure
Load modules/ and helper scripts one phase at a time after the corresponding canaries pass. Preserve typed readiness → Dockerfile → Docker-to-Sealos handoffs, .sealos/analysis.json, .sealos/build/build-result.json, .sealos/template/index.yaml, .sealos/state.json, and delivery evidence; do not hide phase order behind a generic deploy shortcut.
Output, Stop, and Error States
success: actual returned App URL/live identity, route and port match, setup/login proof, recent logs/events, workload convergence, database/object evidence, full footprint, and saved deploy state.
stopped: unsupported eligibility, missing auth/tool, unresolved configuration, or unconfirmed public/destructive/cleanup boundary with the safe next action.
error: failed preflight, build/template/deploy/runtime/rollback/cleanup step and recovery action with sensitive values redacted.
Handoffs
Readiness, Dockerfile, and Docker-to-Sealos inputs use typed target, inputArtifact, allowedAction, failureReturn, and responseOwner fields. A verified deploy can hand target: sealos-canvas, inputArtifact: sanitized .sealos/state.json and Runtime Truth, allowedAction: read-only topology inspection, failureReturn: runtime/state diagnostic, and responseOwner: sealos-deploy.
Verification
Use the existing eligibility, artifact, quality-gate, footprint, live-smoke, rollout, Runtime Truth, and cleanup checks. Baseline cases deploy-positive-runtime-truth and deploy-violating-missing-runtime-proof must preserve actual App URL/live identity, auth/cleanup confirmation, log scans, and redaction.
Compatibility
Sealos auth/workspace are required for deploys. Docker, buildx, and gh CLI are required only when the selected path needs local build/push. git is required when cloning from a GitHub URL or when git metadata is needed. Node.js 18+ remains an optional accelerator. Phase 5 requires Python 3.8+ with PyYAML; root Compose conversion also requires kompose and may require crane when image tags are floating.
Brain Managed Mode
The skill has two deliberately separate execution modes:
Local mode is the existing interactive workflow. It is selected when SEALAI_DEPLOY_MODE is absent or has any value other than managed; its auth, prompts, Template API flow, and output remain unchanged.
Managed mode is selected only when SEALAI_DEPLOY_MODE=managed. The Devbox Codex is the deployment executor: it analyzes, builds, applies, observes, diagnoses, repairs, and verifies with the injected kubeconfig. Brain is the task control plane and form owner; it is not a second Kubernetes executor.
Managed mode is non-interactive. Do not start OAuth, install tools, ask for confirmation in the turn, or replace a missing callback with a file, webhook, curl request, or a Brain-side apply. Before doing any work, confirm that the Codex tool registry contains both exact MCP tools template_ready and deployment_completed. If either tool is unavailable, stop with a managed-mode fatal error; never claim a deployment result.
Brain supplies these task-scoped values through the environment:
Use the injected kubeconfig/context for every Kubernetes command. Do not perform login or switch workspace. Keep the input file and kubeconfig out of prompts, logs, Timeline text, and generated artifacts; read the input file only when the managed flow says to do so.
The actual Sealos Instance name is owned by the Template/Skill path. Brain does not pre-allocate it and the managed adapter does not add Brain identity labels or extraLabels.
Deploy compatible cloud workloads to Sealos Cloud, stopping unsupported targets
before build or deployment.
kubectl Safety Rules (all phases)
All kubectl commands MUST use the Sealos kubeconfig:
System tool installation requires user confirmation. If docker, gh, or kubectl is missing and the skill can install it for the current platform, ask first and only run the install command after the user explicitly replies y.
kubectl delete requires user confirmation. Before deleting any resource (deployment, service, ingress, PVC, database, etc.), always ask:
WARNING: About to delete <resource kind>/<resource name>. This data cannot be recovered. Confirm? (y/n)
Only proceed after user confirms. This applies even if the pipeline logic suggests deletion — always ask first.
Template API cleanup must include Instance CRs. Deployments created through scripts/deploy-template.mjs create instances.app.sealos.io/<app-name> in addition to App/workload resources. A cleanup is incomplete until instances.app.sealos.io, apps.app.sealos.io, workloads, Services, Ingresses, PVCs, and Pods are all checked.
Use this check when cleaning Template API test deployments:
Anti-example: do not report cleanup complete after only checking app,statefulset,svc,ingress,pvc,pod; that misses instances.app.sealos.io/<app-name> and leaves the Sealos Instance layer dirty.
Usage
/sealos-deploy <github-url>
/sealos-deploy # deploy current project
/sealos-deploy <local-path>
Build amd64 image & push to the selected registry (Docker Hub path assumes a public image at deploy time; omitting --registry keeps auto-detect behavior)
Resolve the current region, enforce private sensitive-args files on POSIX, post a local template YAML, and emit an allowlisted result with credential values redacted
Read-only JSON scan of Pod/init/main logs plus Warning Event convergence after readiness, login, and documented API or missing-static-asset checks
sealos-auth.mjs
node sealos-auth.mjs check|login|list|switch
Sealos Cloud authentication & workspace switching
All scripts output JSON. Run via Bash and parse the result.
For public web applications, run sealos-launchpad-network.mjs before HTTP smoke. Acceptance requires ok: true, an open public network, the expected Service port, and an App URL host match. The script emits an allowlisted network summary and excludes raw Launchpad application data, environment variables, Secrets, and kubeconfig content.
Runtime Event acceptance uses two scans. Capture the first report after readiness with no baseline, wait at least 60 seconds, then pass that report through --baseline for the final scan. Extend --min-window-seconds to cover one full known reconciliation, probe, or scheduled-work period. An initial Warning Event is an observation; a Warning that advances after the baseline, an unresolved referenced Secret, a Ready transition, a Pod replacement, or a restart delta is an active failure.
For intentional fault injection, retain a pre-injection report as evidence. After recovery reaches Ready, capture a fresh recovery baseline and compare the final scan against that recovery baseline after the full stability window.
Internal Skill Dependencies
This skill references knowledge files from co-installed internal skills. These are not user-facing — they are loaded on-demand during specific phases.
<SKILL_DIR> refers to the directory containing this SKILL.md. Sibling skills are at <SKILL_DIR>/../:
Confirm the repository root is a supported cloud workload
Any non-eligible result → stop
0.5 — Template Fast Path
Match GitHub repo to a configured Sealos template
No match, or match cannot materialize template YAML
1 — Assess
Clone repo (or use current project), analyze deployability
Score too low → stop
2 — Detect
Route an evidence-confirmed source-ready static tree to the pinned Nginx image build; otherwise find an existing image
Existing image → jump to Phase 5
3 — Dockerfile
Generate Dockerfile if missing
Already has one → skip
4 — Build & Push
docker buildx → GHCR (auto via gh CLI) or Docker Hub (fallback)
—
5 — Template
Generate Sealos application template
—
5.5 — Configure
Guide user through app env vars and inputs
No inputs needed
6 — Deploy
Deploy template to Sealos Cloud
—
6.5 — Runtime Truth Pass
Verify Launchpad public networking, the actual Sealos runtime, logs, Event convergence, App URL, login path, object-storage flow, and resource footprint
User explicitly requests deploy-only output
Decision Flow
Input (GitHub URL / local path)
│
▼
[Phase 0] Preflight ── fail → guide user to fix and STOP
│ pass
▼
[Phase 0.5] Template fast path
│
├── materialized template match ───────┐
│ │
▼ │
[Phase 1] Assess ── not suitable → STOP with reason
│ suitable
▼
[Phase 2] Detect existing image
│
├── found (amd64) ────────────────────┐
│ │
▼ │
[Phase 3] Dockerfile (generate/reuse) │
│ │
▼ │
[Phase 4] Build & Push to registry │
│ │
◄─────────────────────────────────────┘
│
▼
[Phase 5] Generate Sealos Template
◄──────────────────────────────────────┘
│
▼
[Phase 5.5] Configure ── present env vars → ask user for inputs → confirm
│
▼
[Phase 6] Deploy to Sealos Cloud ── 401 → re-auth
│ 409 → instance exists
▼
[Phase 6.5] Runtime Truth Pass ── network/runtime/log/login issue → debug template or runtime config
│
▼
Done — app deployed ✓
Execution rule: Phase 1 must never start while Phase 0 still has unresolved entry blockers. Docker, gh, builder, and registry failures must be reported early, but only become hard blockers if the run later requires local build/push.
