arty-kk/vibe-coding

Use Vibe Coding when the user explicitly invokes the plugin without choosing a specialized skill, asks what it can do, or requests help choosing a workflow. Route by requested outcome and domain; do not trigger for every coding task.

Frame, audit or improve an existing AI feature, RAG pipeline, OCR flow, agent tools, ML pipeline, evaluation, model serving or AI governance. Select the actual model/data/tool boundary.

Audit or fix API contracts and conversational bot flows in an existing repository, including validation, state transitions, callbacks, permissions, and direct consumers.

Audit, fix or verify cache keys, invalidation, TTL, stale reads and coherency across authoritative writes and readers. Focus on cache correctness rather than Redis queue semantics.

Audit, fix or verify Celery task publication, acknowledgements, retries, time limits, prefetch, schedules and warm/cold shutdown against the actual broker and pool.

Audit, fix or verify ClickHouse table engines, partition/order keys, ingestion deduplication, query pruning, merges, replication and workload limits.

Audit, implement or verify a compatibility boundary through authorized reference behavior, file/protocol reconstruction, differential comparison and golden masters.

Audit or edit repository-owned product copy, notification templates, localization, or public SEO/GEO metadata. Ground claims and calls to action in implemented capabilities.

Audit or fix an existing design system or accessible interaction: tokens, shared components, keyboard, focus, semantics and responsive states. Use for design implementation work grounded in code and rendered evidence.

Audit, fix or verify browser extensions or desktop applications: IPC/messaging, permission boundaries, local storage, reload/update, native integration and packaging.

Generate evidence-backed README or design documentation, synchronize stale docs, or plan and implement a scoped migration/refactor. Preserve source ownership and existing compatibility.

Audit, fix or verify repository business contracts: authentication and OAuth/OIDC sessions, RBAC/ABAC, tenant isolation, billing, data models, pipelines, notebook productionization, experiments and product analytics.

Audit, fix or verify feature flag defaults, targeting, backend/UI enforcement, exposure events, kill switches and removal ownership.

Audit, fix or verify Go goroutine or Rust async task ownership, cancellation, pools, backpressure, errors and ordered service shutdown.

Audit, fix or verify agent harness state machines, checkpoint/resume, bounded fan-out, cancellation, durable side effects and replay. This concerns application orchestration code, not creating assistant tasks.

Audit, fix or verify outbound clients, deadlines, retries, trusted proxies, TLS, webhooks and server WebSocket/SSE lifecycle under bounded failure scenarios.

Audit, fix or verify Kafka/Redpanda ordering, producer guarantees, consumer offsets, rebalances, schema compatibility, retry and replay boundaries.

Audit, fix or verify Kubernetes workload rendering, rollout, probes, scheduling, storage, network policy and graceful shutdown for the actual target environment.

Create or refresh evidence-backed repository maps of product capabilities, navigation, owners, data, runtime, permissions, tests, and releases. Use for understanding a codebase or explicitly requested maps.

Audit, fix or verify an MCP server or client: protocol-version compatibility, tool schemas/results, transport, discovery and authorization. Use for MCP integration defects; general agent planning and RAG belong to vibe-ai.

Audit, fix or verify mobile navigation, offline/sync, secure storage, platform permissions, deep links and release configuration for the changed feature.

Audit, fix or verify engine-specific NoSQL keys, conditional writes, atomicity, indexes, hot partitions, TTL, change streams and failover.

Audit, fix or verify notification eligibility, consent, templates, queues, provider status, retry deduplication and delivery recovery. Code work is not authorization to contact real recipients.

Audit, fix or verify Prometheus, Grafana or VictoriaLogs signal semantics, alert rules, dashboards, correlation, tenant scope and cardinality costs in a repository.

Audit or fix runtime configuration, background jobs and CI/CD in a repository. Bootstrap deployment files only for a requested and evidenced deployment target; do not choose a provider by default.

Audit, fix or verify payment provider adapters, signed webhooks, idempotency, event ordering, reconciliation and entitlement consistency. Use sandbox fixtures for payment scenarios.

Run one bounded bug-finding probe in an unvisited backend, frontend, architecture or visual area. Fix one proven defect when requested, retain visible thread history, and stop after one primary area.

Audit or implement a selected product requirement, UI behavior, user journey, onboarding, form, or data table; verify acceptance and metric alignment against an existing repository.

Audit test coverage or release readiness, improve an existing test harness, or verify visual regressions. For review of a concrete diff use vibe-review.

Audit, fix or verify RabbitMQ topology, confirms, routing, acknowledgements, prefetch, dead lettering, connection ownership and redelivery recovery.

Audit, fix or verify client query caches, optimistic mutations, event sequence gaps, WebSocket/SSE reconnect, persisted state and rendering cost in realtime UIs.

Audit, fix or verify artifact provenance, release promotion, rollback compatibility, backup integrity and disaster recovery. Restore and failure drills require an appropriate authorized environment.

Audit, fix or verify Redis Streams, BullMQ or ARQ job identity, locks, acknowledgements, retention, backpressure and recovery. Respect the selected framework delivery semantics.

Audit or fix a concrete reliability, consistency, idempotency, concurrency, hot-path performance or resource-budget problem. Use current code and measurements to identify the limiting owner.

Review an explicit patch, PR, branch or working-tree delta and its direct contract dependencies. Default to findings only; fix review defects when requested. Use for patch review, not whole-repository audits.

Audit, fix or verify document indexing, search ACLs, freshness, ranking/filter behavior, reindexing and failure recovery against product expectations.

Audit, fix or verify a requested application security, abuse-defense, privacy/data-rights or audit-log boundary using reachable evidence. Do not run a broad security audit for ordinary edits.

Audit, fix or verify serverless and edge invocation lifetime, duplicate events, environment reuse and provider runtime limits. Use for existing Lambda/Workers or similar runtime defects; SSR/hydration belongs to vibe-web.

Audit, fix or verify SQL queries, indexes, locks, transactions, migration compatibility, connection pools, replica consistency and failover against the selected engine.

Audit, fix or verify object storage authorization, key/version identity, multipart uploads, checksums, presigned URLs, lifecycle and restore.

Shape implementation tasks, validate scope, reconcile findings, decompose large objectives, or close an iteration. Use for requested planning and synthesis, not as a mandatory step before every edit.

Audit, fix or verify Temporal workflow determinism, Activities, signal/update handling, history bounds, replay and worker version compatibility.

Audit, fix or verify Terraform module contracts, exact plans, state identity, drift, moved/import refactors and recovery. Preparing changes does not itself authorize applying them.

Audit, fix or verify vector embedding/schema compatibility, tenant filters, upserts/deletes, ANN recall, reindex migration and degraded retrieval.

Audit, fix or verify web server/client boundaries: SSR, hydration, serialized data, route caches, server actions and navigation in an existing framework. Use for rendering/runtime defects; product UX and visual design have separate skills.

Normalize evidence, check or refresh stale repository maps, maintain this Vibe Coding plugin, or perform a scoped audit/fix/check when no specialized domain recipe fits.