skills

Plugin skills from marketplace nebius-ps-services in nebius/nebius-ps-services. Commands read from the repo's manifest — 3 days ago.

Skills (75)

• agent-nebius-auth-diagnose

  Read-only diagnosis for current-session Nebius project discovery, agent authentication, quota reads, profiles, selectors, or auth-hook failures. Use when auth is missing/blocked; never mutate auth or IAM.

• agent-nebius-auth-setup

  Use only when explicitly invoked to bootstrap or repair current-project Codex Agent Nebius auth: account/tenant access, credentials/profile, repair lease, or separately requested auth hooks.

• ai-agent-design

  Design provider-neutral production AI agent subsystems: behavior and topology, contracts, tools, context, memory, authority, durability, security, evals, and operations. Use ai-stack when AI technology selection remains, design/app-stack for whole products, and implementation skills for code.

• ai-stack

  Select or review undecided AI components across models, runtimes, training/inference, interoperability, retrieval, evals, safety, and operations. Consume ai-agent-design's frozen agent-subsystem contract; use app-stack for non-AI layers, research for one technology, and design for whole solutions.

• align-skill

  Align existing/scaffolded skills for Agent Skills, Codex, Claude Code, and npx skills installation while preserving functionality, native controls, resources, and evals. Accept local or report-only GitHub sources; use skill-creator for new scaffolds and align for projects.

• align

  Align a project after changes across code, wiring, tests, CI, CLI/help, config, docs, workflows, and skills, with focused quality gates. Use align-skill for skill-folder-only work.

• app-stack

  Select or review an undecided application stack and smallest justified architecture. Route AI layers to ai-stack; not for a fixed stack or isolated implementation.

• apply-security

  Review software, infrastructure, CI/CD, containers, IaC, or design for security risks involving secrets, IAM/auth, exposure, inputs, dependencies, or supply chain. Advise implicitly; patch only when authorized.

• attach-ubuntu

  Use only when explicitly asked for a disposable Ubuntu Docker test environment on macOS: mount the project at /workdir, configure VS Code Dev Containers, and best-effort open a new window.

• brainstorm

  Explore ideas or technical directions in chat before implementation using source-ranked context. Compare options; use research for due diligence and design for decisions/plans; never edit code.

• code-info

  Use only when explicitly asked for a read-only local/GitHub project report: summary, feature/CLI/LOC/dependency counts, tests, artifacts, coverage, size, links, or comparisons. Never modify the project.

• code-review

  Review local diffs/branches, files, modules, or frozen patches for prioritized correctness, regression, test, reliability, and maintainability risks. Standalone may fix safe findings unless report-only; nested PR-diff analysis is read-only. Use review-pr for GitHub orchestration.

• commit-push

  Use when the user asks to commit all repo changes and push an unmanaged non-default branch with repo-root staging and validation. Reject managed worktrees; do not open a PR.

• commit

  Use when the user asks for one local Git commit without push: inspect the full diff, stage repo-wide, validate, commit, and report. Also handles exact worktree-integration commits; not for PRs.

• config-claude

  Use only when explicitly asked to configure, reconcile, inspect, or recover a personal Claude Code setup: native settings, instructions, roles, hooks, private task state, and selected MCP integrations.

• config-codex

  Use only when explicitly asked to configure or recover a public-safe Codex home: config, global policy, MCP, hooks, task state, read-only agents, workspace access, and validation. Never copy private state or secrets.

• container

  Create, review, harden, optimize, or troubleshoot OCI images and container stacks: Dockerfiles, buildx, Compose, runtime/security, multi-platform/GPU, and SBOM. Use sibling skills for app design, CI, publishing, or Helm/Kubernetes.

• create-learning-course

  Use only when explicitly invoked to create or revise public-safe courses, syllabi, lessons and practical labs with definition-first teaching and a self-contained HTML textbook. Not for live tutoring, product docs, LMS deployment or publishing.

• create-pr

  Use only when explicitly asked to create a GitHub PR from unmanaged work: validate, commit, code-review before publication, push, and verify PR readiness. Stop at the reviewed PR by default; explicit merge intent delegates merge-pr. In active Agentic SDLC, publish only the exact promoted SHA after U

• design

  Design non-SDLC features, architectures, ADRs, proven contract-changing remediations, or standalone portable HTML artifacts. Ground plans in code and docs; implement only on explicit request. Use research for due diligence, ai-agent-design for AI behavior, app-stack/ai-stack for stack choices, and n

• frontend-project

  Create or standardize React, TypeScript, and Vite frontends/components with strict config, routes, environment checks, tests, tooling, docs, and candidate manifests. Not for Node backends, other frameworks, Sites, containers, root CI, or stack selection.

• github-report

  Report GitHub activity grouped by human contributor and one shared open backlog. Accept --user login/name substring, --date, comma-separated --repo and folder filter --project; default to all users, current repo and last 90 days. Read-only reporting, not issue triage, PR code review or contributor r

• github-workflows

  Create, review, or standardize GitHub Actions for PR/merge CI, bot-safe automation, reusable workflows, permissions, scoped checks, or release/image jobs. Use publish-* for complete publication flows.

• gitignore

  Create, fix, or standardize a project's .gitignore with macOS/VS Code defaults and detected stack rules for Node, Python, Go, Rust, Java, Terraform, or others.

• global-context-management

  Manage complex agent planning, implementation, debugging, refactoring, migration, review, testing, CI, or multi-file work with concise parent context, durable task state, bounded read-only helpers, focused validation, and risk review.

• helmchart

  Create, review, harden, refactor, lint, or standardize Helm charts, values/schema, templates, NOTES, README, and chart CI. Use publish-helm for OCI publication; not for plain Kubernetes, Dockerfiles, or Terraform.

• implement-test

  Use only when the user explicitly asks to verify Implement: run the no-flag contract/fixture suite, or use --create/--keep/--destroy for one replaceable verifier-owned local frontend/API/PostgreSQL stack.

• implement

  Use only when explicitly invoked to implement the current plan or an inline request with isolated task workers, verified local delivery and durable continuation. Use continue for a new session; not SDLC or remote publication.

• install-grafana-mcp-for-nebius

  Use only when explicitly invoked. Install, configure, validate, or repair Grafana MCP for Nebius-managed Grafana with a pinned human profile, Codex config, token wrapper, or requested datasources. Use nebius-grafana-query for queries.

• jira-report

  Report a person's Jira involvement through an existing MCP connection, including assignments, Participants, reporting, comments and credited contributions. Accept --user, --date and --title; default to self and 90 days. Read-only reporting, not ticket editing, triage or connector setup.

• linter

  Lint and auto-fix repository Shell, Markdown, or Python files using shellcheck/bash -n, markdownlint with project fallbacks, or Ruff with targeted pyproject rules.

• maintain-project-specs

  Maintain canonical requirements and design when direct root-user intent or proven delivery changes project truth, across ordinary sessions, Implement, and Agentic SDLC. Hooks never gate tools, Stop, or workflow completion.

• merge-pr

  Use only when explicitly asked to complete an ordinary GitHub PR via local review-pr and a protected GitHub Actions merge; verifies exact head, checks and result. Agentic SDLC is excluded.

• nebius-audit-log

  Use only when the user explicitly asks to investigate or query tenant-level Nebius Control Plane Audit Logs by resource, actor, service, action, status, region or time. Verifies caller and audit access; not for general cloud logs or auth repair.

• nebius-capacity-planning

  Size Nebius compute, storage and networking from project designs, using conversation or local design documents, component allocations, headroom, all-region tenant checks and an auditable report. Optional quota requests require explicit authority. Use nebius for inspection alone and terraform/helmcha

• nebius-grafana-query

  Run read-only Nebius Grafana MCP queries for scoped dashboards, PromQL, Loki, or Tempo, including troubleshoot/SDLC evidence. Never install or configure MCP, repair auth/IAM, query audit logs, diagnose causes, grade acceptance, or mutate Grafana.

• nebius

  Design, build and inspect Nebius infrastructure across compute, storage, networking, observability, IAM and API/SDK. Advise design on provider constraints without credentials or cloud actions. Route Terraform, Helm, cxcli lifecycle, Grafana queries and agent-auth setup to their owners.

• nosleep4mac

  Use only when the user explicitly requests a per-user macOS LaunchAgent that keeps the Mac awake on AC with caffeinate -s. Not for temporary sessions, display/battery policy, closed-lid promises, or system daemons.

• optimize-pytest

  Measure and safely optimize slow pytest suites by separating startup, collection, fixture, call, and teardown costs and proving like-for-like gains. Not for generic failures, flakiness, scaffolding, or production performance.

• project-agent-instructions

  Explicitly repair restrictive project instructions, or render, create, refresh, adopt, or retire selected-project AGENTS.md rules with receipt-bound ownership and recovery.

• publish-helm

  Use only when explicitly asked to publish Helm charts end to end: collect chart/OCI inputs, set up optional assets, prepare/merge a PR, tag, wait, verify the OCI chart, and report. Also supports setup-only guidance.

• publish-image

  Use only when explicitly asked to publish container images end to end: collect inputs, set up optional assets, prepare/merge a PR, tag, wait, verify tags/digest, and report. Use github-workflows for workflow YAML.

• publish-release

  Use only when explicitly asked to publish a GitHub Release: commit current work, push and merge its PR, locally review/fix, use protected Actions merge, tag, verify assets, and resume interruptions. Also supports setup-only guidance.

• python-project

  Own Python candidates and readiness for scaffold-project creation; directly maintain Python packaging, dependencies, tests, and tooling with uv, src layout, Ruff, and pytest. Preserve an existing package manager unless uv migration is requested; route domain infrastructure elsewhere.

• research

  Research one technical topic, technology, pattern, product, framework, RFC, protocol, API, requirement, or disputed claim using ranked authoritative sources. Not for implementation, brainstorming, stack selection, final design, or SDLC artifacts.

• review-github-issues

  Use only by explicit invocation to review GitHub issues: discover open backlogs or select issue numbers across repositories, group by project and shared remediation, apply justified triage, and save batch plans. Never implements work or reviews PRs.

• review-pr

  Use only when explicitly asked to review a GitHub PR by number, URL, or branch: coordinate code-review, safe repairs through create-pr, GitHub checks and readiness. Plain review never merges; explicit combined review-and-merge delegates once to merge-pr. In Agentic SDLC, report only.

• scaffold-project

  Coordinate creation of repositories and components, including Python-only projects, with specialist-owned candidates, idempotent apply, and Python readiness. Use for scaffold creation or topology changes; route existing-project maintenance to its specialist. Not for stack selection, feature logic, b

• sdlc-auto-steering

  Use only as part of the Agentic SDLC workflow; refresh private STEERING.md for an active objective-bound run from accepted revisions, specs, context, locked plans, fingerprints, or evidence before phase selection.

• sdlc-classify-failure

  Use only as part of the Agentic SDLC workflow; classify a failed phase or evaluation with deterministic evidence, diagnosis, repair budget, invalidation, and the earliest responsible retry or stop route.

• sdlc-commit

  Use only as part of the Agentic SDLC workflow; after waves/evidence pass, seal integration changes, fast-forward the unchanged project branch to the exact tip, and non-force-clean the integration resource. Never push.

• sdlc-create-design

  Use only as part of the Agentic SDLC workflow; adapt requirements and evidence into canonical FEAT records in docs/design.md through maintain-project-specs, preserving stable IDs and shared-owner validation.

• sdlc-create-plan

  Use only as part of the Agentic SDLC workflow; lock a dependency-safe plan for one admitted delivery slice, or append immutable corrective waves in plan vN+1 after a classified evaluation failure.

• sdlc-create-requirements

  Use only as part of the Agentic SDLC workflow; adapt user intent into canonical REQ records in docs/requirements.md through maintain-project-specs, preserving stable IDs and shared-owner validation.

• sdlc-evaluate

  Use only as part of the Agentic SDLC workflow; after validation/tests, judge whether a feature and end-to-end slice meet acceptance criteria using the right harness, safe live trials when needed, and predefined observability evidence.

• sdlc-gather-context

  Use only as part of the Agentic SDLC workflow; gather technical, product, vendor, internal, codebase, layer, and test evidence into a compact feature-specific context pack before design or implementation.

• sdlc-gui-test

  Use only as part of the Agentic SDLC workflow; control, observe, screenshot, and evaluate browser GUI behavior through the declared harness, using headless Playwright Test for web acceptance and optional isolated Playwright MCP exploration.

• sdlc-implement-plan

  Use only as part of the Agentic SDLC workflow; after sdlc-tdd, coordinate one delivery slice's dependency waves with isolated agents, branches, and worktrees, ordered integration, combined validation, and non-force cleanup.

• sdlc-merge-pr

  Use only as part of the Agentic SDLC workflow; complete an opted-in delivery slice through protected direct merge and exact-result verification after independent qualification, review and policy admission.

• sdlc-prepare-execution

  Use only as part of the Agentic SDLC workflow; after a feature plan is locked and before sdlc-tdd, create or resume its private integration branch/worktree, deterministic task waves, and recoverable execution state.

• sdlc-start

  Use only as part of the Agentic SDLC workflow; explicitly start from a direct objective or current plan, continue the same session's bound run, or explicitly continue a project across sessions. Own phase routing, checkpoints, accepted steering and completion.

• sdlc-tdd

  Use only as part of the Agentic SDLC workflow; after sdlc-prepare-execution, write feature tests in the registered integration worktree before dependency-wave implementation.

• sdlc-tui-test

  Use only as part of the Agentic SDLC workflow; control, observe, transcribe, and evaluate terminal, CLI-wizard, or TUI behavior against acceptance criteria with Pexpect-style interaction.

• sdlc-uat-tests

  Use only as part of the Agentic SDLC workflow; qualify the promoted delivery slice and inherited product regression obligations before PR creation, using a confirmed safe live target when applicable.

• sdlc-unit-tests

  Use only as part of the Agentic SDLC workflow; after validation, run applicable unit, integration, component, contract, regression, and mock behavior tests for the feature and planned end-to-end slice.

• sdlc-update-documents

  Use only as part of the Agentic SDLC workflow; update evaluated feature documentation before promotion, assess later documentation read-only, and route required corrections without editing requirements/design.

• sdlc-validate-codes

  Use only as part of the Agentic SDLC workflow; after implementation or before sdlc-commit, run build, parse, lint, type, import, dependency, config, and locked-slice checks, then a report-only code-review gate.

• sdlc-workflow-test

  Use only when explicitly asked, outside Agentic SDLC, to verify the workflow with the no-flag check or --create-live-test/--keep/--resume/--destroy with deterministic checks, one owned local three-tier Docker app, GUI UAT and cleanup of both fixture types.

• shell-scripting

  Create, refactor, review, or harden Bash scripts, shell automation, and CLI wrappers with strict mode, safe parsing, idempotency, structured logs, readable help, and color fallback.

• system-design-rules

  Review architectures, ADRs, boundaries, APIs, data ownership, reliability, security, observability, scaling, cost, and team ownership with a 100-rule checklist before implementation.

• terraform

  Scaffold, standardize, or harden Terraform repos/modules: structure, state/backends, environments, interfaces, tests, security, and CI. Not for nebius-cxcli wizard/status/catalog work unless changing Terraform modules.

• test-software-in-container

  Test the current project source live in disposable containers, or explicitly test installed images, installations, upgrades, patches and migrations. Not VM, hardware or production qualification.

• troubleshoot

  Diagnose and repair difficult, persistent, intermittent, or cross-layer failures in code, compute, storage, and networks through bounded experiments and causal proof. Not for routine lint, setup, review, feature work, or known mechanical fixes.

• update-branch

  Use only when the user explicitly asks to update the current local feature branch from origin's default branch. Merge in place in a clean ordinary checkout; no branch switching, rebase, push, or feature-upstream sync.

• worktree

  Requires explicit invocation to add, integrate, reuse, or remove full-repo linked Git worktrees with exact branch, commit, and cleanup checks. Not for push, PRs, publication, or parallel-agent orchestration.

Hooks (10)

• PreToolUse

  python3 "${CLAUDE_PLUGIN_ROOT}/global-context-management/scripts/hook_runtime.py" --agent claude --hook pre_tool_use_nebius_auth.py --plugin-root "${CLAUDE_PLUGIN_ROOT}"

• PreToolUse

  python3 "${CLAUDE_PLUGIN_ROOT}/global-context-management/scripts/hook_runtime.py" --agent claude --hook pre_tool_use_sdlc_policy.py --plugin-root "${CLAUDE_PLUGIN_ROOT}"

• PreToolUse

  python3 "${CLAUDE_PLUGIN_ROOT}/global-context-management/scripts/hook_runtime.py" --agent claude --hook remediation_attempt_guard.py --plugin-root "${CLAUDE_PLUGIN_ROOT}"

• UserPromptSubmit

  python3 "${CLAUDE_PLUGIN_ROOT}/global-context-management/scripts/hook_runtime.py" --agent claude --hook commit_intent.py --plugin-root "${CLAUDE_PLUGIN_ROOT}"

• UserPromptSubmit

  python3 "${CLAUDE_PLUGIN_ROOT}/global-context-management/scripts/hook_runtime.py" --agent claude --hook user_prompt_context.py --plugin-root "${CLAUDE_PLUGIN_ROOT}"

• UserPromptSubmit

  python3 "${CLAUDE_PLUGIN_ROOT}/global-context-management/scripts/hook_runtime.py" --agent claude --hook project_specs_lifecycle.py --plugin-root "${CLAUDE_PLUGIN_ROOT}"

• UserPromptSubmit

  python3 "${CLAUDE_PLUGIN_ROOT}/global-context-management/scripts/hook_runtime.py" --agent claude --hook remediation_attempt_guard.py --plugin-root "${CLAUDE_PLUGIN_ROOT}"

• SessionStart

  python3 "${CLAUDE_PLUGIN_ROOT}/global-context-management/scripts/hook_runtime.py" --agent claude --hook session_start_context.py --plugin-root "${CLAUDE_PLUGIN_ROOT}"

• SessionStart

  python3 "${CLAUDE_PLUGIN_ROOT}/global-context-management/scripts/hook_runtime.py" --agent claude --hook project_specs_lifecycle.py --plugin-root "${CLAUDE_PLUGIN_ROOT}"

• Stop

  python3 "${CLAUDE_PLUGIN_ROOT}/global-context-management/scripts/hook_runtime.py" --agent claude --hook stop_lifecycle_arbiter.py --plugin-root "${CLAUDE_PLUGIN_ROOT}"

Similar plugins