mit einem Klick
mcp-runtime
mcp-runtime enthält 18 gesammelte Skills von Agent-Hellboy, mit Repository-Berufsabdeckung und Skill-Detailseiten auf SkillsMP.
Skills in diesem Repository
Real-cluster security regression QA — backend auth enforcement, grant/session policy, gateway deny paths, audit emission, trust escalation, UI/API security headers, HTTPS redirect modes, and secret-leak scanning in live logs. Use when Codex is asked to verify a change did not regress auth, governance, gateway policy, audit, or UI security headers on a live cluster; or for pre-release security-regression sweeps. Complements static-only security-audit/security-audit-platform with real traffic. Assumes qa-cluster-bringup has run.
Run a deep platform-wide security audit of MCP Runtime — every component, every trust boundary, against an explicit threat model. Use when Codex is asked for a thorough security review (not a single PR), pre-release sign-off, red-team-style assessment, or to produce a structured report of findings across operator, gateway, sentinel services, registry, ingress, governance, and CI. For change-scoped reviews use security-audit; for SBOM/image/dependency work use supply-chain-audit; for cluster RBAC/PSS/NetworkPolicy hygiene use k8s-hardening-audit.
API and system design principles for MCP Runtime — RESTful resource design, HTTP semantics, versioning, and the broader system principles (least privilege, secure/deny by default, fail-closed, stable contracts, separation of concerns, validate at boundaries, idempotency, observability, audit, minimized blast radius, good defaults). Use when designing or reviewing an API, CRD, CLI flag, operator/gateway/service behavior, auth/policy path, or any new contract; and as the review lens for "is this good design?".
Real-cluster operational QA for MCP Runtime — operator/CRD reconciliation, CLI flows, setup/test-mode regressions, registry pulls, ingress wiring, rollout health, and observability — against the live Kind contributor cluster. Use when Codex is asked to verify a change does not regress operator, CLI, setup, registry, ingress, or rollout behavior; to do release-readiness operational QA; or to investigate a real-cluster failure. Assumes qa-cluster-bringup has run.
Compose MCP Runtime pre-merge, ship, canary, and release readiness across CI parity, live Kind operations, security, UI/browser QA, protocol compliance, performance, docs, and deployment evidence. Use when asked if a branch can ship, to prepare a release, run a canary, produce release evidence, or coordinate final review before merge/tag/deploy.
Operate or debug the MCP Runtime public k3s deployment and related hack scripts, including setup, clean/restore, rollout, registry TLS/auth, ImagePullBackOff, node DNS vs pod DNS, and post-change live validation. Use when touching hack/deploy/mcpruntime-org/, config/deployments/mcpruntime-org.env.example, public TLS registry behavior, or k3s deployment docs.
Audit MCP Runtime Kubernetes posture — RBAC, ServiceAccounts, Pod Security Standards, NetworkPolicies, manifest hygiene, secret access, and cluster CIS controls — across operator, sentinel, registry, traefik, and mcp-servers namespaces. Use when Codex reviews or changes RBAC roles, ServiceAccounts, Deployments, NetworkPolicies, PSS labels, securityContext, or k8s manifests under k8s/, config/, or service charts; or when asked for a cluster hardening review.
Apply and debug MCP Runtime access grants, agent sessions, gateway policy, and MCP JSON-RPC traffic with governance headers. Use when working on MCPAccessGrant, MCPAgentSession, adapter proxy/stdio, access CLI, platform API grant/session endpoints, or allow/deny tool calls.
Local Kind contributor endpoints, API keys, test-mode logins, port-forward URLs, and Sentinel auth for MCP Runtime. Use when developing against kind-mcp-runtime, curling `/api/v1` or MCP paths on localhost:18080, or debugging 401s in test-mode — after cluster bring-up (qa-cluster-bringup).
Configure and debug public MCP Runtime installs with MCP_PLATFORM_DOMAIN, Let's Encrypt, registry/mcp/platform hostnames, registry TLS auth, and platform UI ingress. Use for --with-tls setup, ACME failures, DNS NXDOMAIN, registry forward-auth, or production hostname routing — not for local Kind test-mode (use qa-cluster-bringup).
Debug MCP Runtime cluster, ingress, registry, Sentinel auth, MCPServer pods, and platform UI failures on Kind or k3s. Use when setup, doctor, e2e, or live traffic fails; when investigating 401/404/ImagePullBackOff/ACME/TLS/registry push errors; or when AGENTS.md points here for the full failure-mode checklist.
Stand up a real Kind+test-mode MCP Runtime cluster end-to-end (the contributor flow from docs/getting-started.md#3-contributor-test-mode-cluster) so the qa-e2e-* skills can run real validation. Use when Codex is asked to QA on a live cluster, prepare a target environment for qa-e2e-operations / qa-e2e-security / qa-e2e-ui / qa-e2e-perf, recover a broken contributor cluster, or verify a fresh setup --test-mode end-to-end. Owns cluster lifecycle; the other QA skills assume this has run.
Real-cluster performance regression QA — MCP tools/call latency (p50/p95/p99), concurrent throughput against the gateway, /api/v1/analytics/usage latency under load, and operator burst-reconcile latency. Compares against a stored per-branch baseline and flags regressions above a configurable threshold (default +25 percent on p95). Use when Codex is asked to perf-check a change to gateway/proxy/operator/api hot paths, before a release, or when investigating a "feels slower" report. Assumes qa-cluster-bringup has run.
Browser-first real-cluster Sentinel UI/dashboard QA - role-based navigation, auth flows, every tab, forms, filters, destructive actions, rendered data, network/API evidence, console evidence, responsive/accessibility checks, cleanup, static assets, and public-host defenses. Use when Codex is asked to QA UI changes, dashboard regressions, login/admin/tenant flows, browser-visible API behavior, copyable MCP connect config, or backend analytics/observability changes that must be validated through UI controls. Complements qa-e2e-security with feature-correctness and browser interaction checks. Assumes qa-cluster-bringup has run.
Audit a single MCP Runtime change for security risks and choose the relevant local or CI security checks. Use when Codex reviews or changes authentication, authorization, secrets, TLS, ingress, registry/image handling, Kubernetes RBAC or pod security, gateway policy, audit logging, dependencies, GitHub Actions, container builds, or any code path that accepts untrusted input. For a deep platform-wide audit prefer security-audit-platform; for SBOM/image/dependency-only audits prefer supply-chain-audit; for cluster RBAC/PSS/NetworkPolicy hygiene prefer k8s-hardening-audit.
Audit MCP Runtime supply chain — Go modules, container images, base images, SBOMs, signatures, and GitHub Actions workflows — for known vulnerabilities, unsigned artifacts, weak action pinning, license violations, and pwn-request exposure. Use when Codex bumps dependencies, changes Dockerfiles or workflows, prepares a release, or is asked to audit dependencies, images, SBOMs, signatures, or CI security.
Audit and update repository documentation, AGENTS.md, contributor runbooks, and test guidance so they match implemented behavior. Use when code, UI, CLI, API/CRD, config, deployment, or test-flow changes may affect user-facing docs, setup instructions, operational/debugging guidance, golden help snapshots, E2E assumptions, or agent instructions; when the user asks to sync docs after implementation; or when stale repo guidance could mislead contributors. Do not use for ordinary code review or implementation work unless documentation or guidance drift is part of the task.
Validate MCP Runtime against the upstream Model Context Protocol specification at github.com/modelcontextprotocol/modelcontextprotocol — protocol version negotiation, Streamable HTTP transport headers, JSON-RPC 2.0 envelope, initialize/tools/resources/prompts/sampling shapes, session semantics, error codes, and merged + open SEPs. Runs static checks against the runtime code and dynamic checks against the live Kind cluster. Use when Codex is asked to validate MCP compliance, audit transport/protocol behavior, prepare for an upstream protocol bump, or compare runtime behavior against a specific spec revision. Complements qa-e2e-security (auth/governance) and qa-e2e-operations (operator/CLI) with protocol-level conformance.