Claude Skills by atretyak1985

Generate or refresh the repo-wide architecture map — architecture-out/architecture-map.json (machine contract with named flows) + architecture-map.html (self-contained viewer). Use when the user asks for an architecture map, repo map, "/architecture-map", or when an agent needs a fresh machine-readable architecture overview. NOT for per-epic C4 deep-dives (use c4-architecture-docs) and NOT for building the knowledge graph itself (use /graphify).

Use when designing or reviewing Claude-based agent systems — agentic loops and stop_reason handling, multi-agent orchestration, coordinators, subagent invocation and context passing, workflow enforcement, Agent SDK hooks (PreToolUse/PostToolUse), task decomposition, or session state and resumption. Also when an agent terminates prematurely, loops forever, repeats work, or subagents return inconsistent or incomplete results. NOT for tool schemas (use designing-tools-and-mcp), prompts (use engi...

Use when setting up or auditing Claude Code in a repository — CLAUDE.md hierarchy and imports, .claude/rules, custom slash commands and skills, path-specific rules, plan mode vs direct execution, iterative refinement workflow, or CI/CD integration with headless -p mode. Also when instructions get ignored, conventions load in the wrong scope, CLAUDE.md has grown bloated, or a CI pipeline needs non-interactive Claude runs. NOT for task prompt content (use engineering-prompts-and-output), agent ...

Use when creating or reviewing tool definitions for Claude — tool names, descriptions, input schemas, structured error responses, tool_choice configuration, MCP server setup and scoping, or built-in tool usage (Grep, Glob, Read, Edit). Also when Claude picks the wrong tool, confuses similar tools, retries hopeless failures, treats empty results as errors, or degrades because too many tools are exposed. NOT for agent loops/orchestration (use auditing-agent-architecture), system prompts/schemas...

Use when writing or reviewing system prompts, few-shot examples, structured output via tool use, JSON schema validation with retry loops, Message Batches API processing, or multi-instance/multi-pass review pipelines such as automated PR reviewers. Also when outputs are inconsistent between runs, JSON fails validation, false positives erode reviewer trust, hallucinated fields appear, or self-review keeps missing defects. NOT for tool interface design (use designing-tools-and-mcp), context/revi...

Use when reviewing long-running or multi-agent Claude systems for reliability — context window management and summarisation, escalation and ambiguity resolution, error propagation between agents, codebase exploration and context degradation, human review and confidence calibration, or information provenance and multi-source synthesis. Also when quality degrades over long sessions, agents silently swallow failures, mid-context details get lost, or synthesised answers lose attribution. NOT for ...

Verify field-level alignment between ORM schema, Zod schemas, and route handler shapes, or that SQL migrations match ORM tables. NOT for writing new routes (use api-integration) or code style (use code-standards).

Implement REST route handlers, ORM queries, WebSocket telemetry, or SSE endpoints in the main web app, or consume those APIs from clients. NOT for schema-only changes or deployment/infra config.

Convert an operational runbook (pod restarts, cache flushes, scaling) into a parameterized idempotent script with safety gates, or design a chaos experiment. NOT for CI/CD or deployment manifests (use deployment).

Verify UI behavior in a live browser via Playwright MCP (navigate, snapshot, screenshots, console/network) against localdev or staging. NOT for full domain E2E flows and never against production.

Document an epic/feature architecture with the C4 model -- system context/container/component/dynamic diagrams as Mermaid .mmd plus a narrative doc. NOT for rendering an existing .mmd (use mermaid-viewer).

QUANTITATIVE structural metrics on TS/Python -- function length, cyclomatic complexity, nesting, duplication, code smells -- scored report. NOT for conventions/`any` checks (use code-standards) or API alignment (use api-contract).

Select the right search tool (Grep, Glob, codebase-retrieval) to locate existing code, symbols, or files across the project's monorepo. Don't use it for creating new files, implementing features, or reviewing code quality.

Review code against CONVENTIONS -- type safety (`any` detection), naming, imports, Next.js patterns, 12-factor. NOT for complexity metrics (use code-quality) or API field alignment (use api-contract).

Use this skill when a task spans 3+ files or crosses repo boundaries and context usage must be managed. Don't use it for single-file edits or read-only queries.

Use this skill when auditing dependency versions, checking for outdated packages, or scanning for security vulnerabilities across the project's repos. Don't use it for upgrading packages (that requires a separate implementation task) or for deployment config template issues (use deployment).

Use this skill when building or pushing Docker images for the project's services (the main app, the device service) to the cloud container registry. Don't use it for deploys (use deployment) or Dockerfile template editing without a build (use code-quality).

Use this skill when a task involves adding, removing, or renaming environment variables across the project's repos OR verifying env var documentation before a release. Don't use it for runtime env introspection in a live cluster (that requires exec access to the running service).

Use this skill when the user asks for an investor-mode reality check on the product/business — audit the shipped state against the business thesis with VC-style critique. Don't use it for code review or feature planning.

EXECUTES refactors of TypeScript business logic toward pure functions, immutability, and pipelines -- edits files via Edit. NOT for planning-only refactors (use refactor-plan) or I/O-heavy handlers / Python paths.

Generate conventional commit messages when staging files for commit in any of the project's repos. Do not use for git tag messages, merge commit messages, changelog entries, or automated version bump commits.

Use this skill before executing a risky or hard-to-reverse action — migrations, deletions, deploy/config changes, force operations — to derive a risk level and an APPROVED/REJECTED call with a rollback plan. Don't use it for routine read-only work.

Render a self-contained HTML report or dashboard (task summaries, code/operational audits) in one canonical dark-terminal shell. NOT for markdown-only output or the mermaid viewer (use mermaid-viewer).

Read-only Jira for the project's tracker: my tickets, open tickets, ticket status, jira backlog, link a workspace task to a ticket. NOT for write ops (create/transition/comment/log-work) and NOT for Confluence.

Use this skill when the user has a Mermaid source file (.mmd) and wants to view it in a browser -- asking to convert mermaid to html, render this diagram, make an HTML viewer, or make the schema explorable. Don't use it for generating new diagrams from scratch or exporting to PNG/PDF.

Validate SQL migration safety, ORM-to-migration alignment, or schema drift between the DB and Zod/DTO types. NOT for deployment config (use deployment) or IaC drift (use infrastructure-as-code).

Prometheus metrics, Grafana dashboards, alert rules, ServiceMonitor wiring, and endpoint instrumentation. NOT for logs/traces (belong to observability) and NOT for Helm health probes.

Coordinate changes spanning 2+ repos of a multi-repo workspace or 2+ apps/packages of a monorepo -- merge-order plans, MR/PR templates, post-merge checklists. NOT for single-repo changes, even large ones.

Structured logging and OpenTelemetry distributed tracing, incl. log-trace-metric correlation. NOT for metrics/dashboards/alerts (belong to monitoring) and NOT for Helm health probes.

Use this skill when creating a pull request — generate the title, structured description, and review checklist from the branch's actual commits and diff. Don't use it for commit messages (git-commit skill) or changelogs.

PRODUCES a refactoring plan ONLY -- impact analysis, step ordering, risk, rollback -- NO code changes. NOT for executing refactors (use functional-design) or dependency upgrades (use deps-check).

EXECUTES an existing @planner plan -- parses the phase DAG, dispatches per-phase implement+review loops, preserving ASK gates. NOT for creating plans; NOT usable from inside a subagent.

Scan application code for security vulnerabilities, OWASP Top 10 compliance, or hardcoded secrets across the project's repos. NOT for CI/CD hardening or SBOM generation (use supply-chain-security).

Use this skill when closing a finished task — writing the SUMMARY.md final report, the phase-09 retrospective, and task documentation in the workspace task dir. Don't use it for mid-task status updates or commit messages.

Use this skill for an end-of-sprint read-only audit — scope recent changes across repos, fan out review agents, run check-only suites, and produce a PASS/FAIL report. Don't use it for reviewing a single PR or diff.

Use this skill for SRE work — SLO definition, incident response, capacity planning, toil reduction — with human gates on every destructive operation. Don't use it for application debugging (troubleshooting skill) or metrics wiring (monitoring skill).

Use this skill when a task involves summarizing completed work, writing a feature summary, documenting what changed, or creating a completion report for a task, feature, bug fix, or refactoring. Don't use it for writing documentation, changelogs, or postmortems.

Audit and harden container supply-chain controls: image scanning, SBOM generation, immutable digest promotion, rollback retention, and image signing readiness. Not for application code vulnerabilities (use security-audit) or pipeline YAML structure (use deployment).

FIND test coverage gaps (read-only) -- map source to test files, flag untested modules by risk. NOT for writing/running/debugging tests (use testing) or measuring coverage percentages (run --coverage directly).

WRITE, RUN, or DEBUG tests -- pytest, Jest/RTL, Playwright E2E, deployment config. NOT for finding coverage gaps (use test-coverage) or CI pipeline config (use deployment).

Debug a specific failure, investigate an incident, analyze error logs, or diagnose connectivity on the project's platform. NOT for proactive instrumentation (use monitoring/observability) or writing tests (use testing).

Use this skill when writing, revising, or executing a workspace plan — the plan/README.md + phase-N-<slug>.md format the dashboard ingests. Don't use it for ad-hoc todo lists or in-repo design docs.

Resolve a design handoff into something measurable: parse a handoff prompt, unpack an exported HTML/zip, pull tokens from a design-system source, or record an honest degraded mode for screenshots-only input. NOT for the implementation workflow itself (that's design-implement) and NOT for measuring a diff (that's design-verify).

Use when a finished visual design has to be re-expressed in this project's stack pixel-accurately — the user says \"implement this design\", \"here's a design handoff\", \"make this screen match the mockup\", \"design-implement\", pastes a design-tool handoff prompt, or attaches an exported design HTML/zip. Runs a measured workflow: token inventory from a real headless render, reuse-vs-create recon against the project's own components, an approval gate before any code, and a pixel diff agains...

Measure an implemented screen against its design export with a headless pixel diff and report the regions, the number and the side-by-side image. NOT for choosing what to build (that's design-implement), NOT for resolving the design input (that's design-acquire), and NOT for behavioural UI checks without a reference design (that's browser-verification in core).

Use for any question about a codebase, its architecture, file relationships, or project content — especially when graphify-out/ exists, where the question should be treated as a graphify query first. Turns any input (code, docs, papers, images, videos) into a persistent knowledge graph with god nodes, community detection, and query/path/explain tools. NOT for direct symbol searches (use code-search) or rendered diagrams (use architecture-map).

Configure and review AWS auth for GitHub Actions CI/CD — OIDC role assumption, ECR push perms, ECS deploy perms, least-privilege IAM. Not for application IAM/Cognito, not for GCP.

Use this skill for a Helm-upgrade deploy of an app service (the web portal or the edge service) to an EXISTING Kubernetes cluster, plus post-deploy verification. NOT for cluster-level ops -- Minikube/k3s lifecycle, GCP firewall, tunnels, bootstrap secrets (use kubernetes-deployment); NOT for cross-environment promotion (use release-promotion).

Configure and review GCP authentication for GitLab CI/CD pipelines: Workload Identity Federation, Artifact Registry push, Secret Manager access, and least-privilege service accounts. Not for Keycloak IAM, AWS/Azure auth, or runtime pod identity.

Design, review, or debug GitHub Actions workflows — CI jobs (type-check, lint, test, build, security, summary), AWS ECR/ECS deploys, and a standardized per-repo workflow set. NOT for cloud-auth role/OIDC trust details (use aws-cicd-auth) or Dockerfile/image-build mechanics (use docker-build).