vibecoded-orchestrator

Automated visual testing with Playwright MCP - test web apps, presentations, websites, and documents with scalable reviewer perspectives

Expert guidance on RAG (Retrieval-Augmented Generation) system design including chunking strategies, embedding selection, retrieval methods, and vector database choices

Design complex system architectures, evaluate tradeoffs, and make critical technical decisions requiring deep reasoning

Deep code analysis identifying subtle bugs, security issues, performance problems, and architectural concerns requiring expert-level reasoning

Efficient context state inspection, task lifecycle management, and session tracking

Systematic codebase onboarding. Builds a mental model of a new or unfamiliar project by exploring structure, architecture, key data models, entry points, and auth patterns.

Systematically extract knowledge from scattered documentation to prevent catastrophic forgetting. Creates structured extraction reports with status tags.

Investigate and fix a GitHub issue or bug report. Reads the issue, reproduces it, identifies root cause, implements fix, adds regression test.

Research using ONLY knowledge graph semantic search (no file tools, forces KG-first approach)

Test-Driven Development workflow. Write failing test first, then implement to make it pass. Use for new features, bug fixes, or refactoring with confidence.

Analyzes project workflow setup and suggests/creates needed automation for hooks, scripts, skills, and agents

Quick A11y review - WCAG 2.1 checklist, screen reader compatibility, keyboard navigation, color contrast

Crafts production-grade prompts for AI image generation (Midjourney, Flux, SDXL, Firefly, Imagen, ComfyUI workflows) — subject, composition, lighting, style references, negative prompts, ControlNet hints. Use when the goal is on-brand, repeatable imagery rather than a one-off lucky generation.

Quick guidance on choosing AI models (LLM/VLM/Embedding) based on task, VRAM, cost, and quality requirements

Quick tips and templates for effective prompt engineering - few-shot examples, chain-of-thought patterns, constraint specification, output formatting

[PRE-ALPHA] Generate a Mermaid flowchart of a code subgraph centered on a function, class, or module. Runs `vco codegraph-diagram` under the hood, writes the .mmd to .claude/diagrams/codegraph/ so it's indexed for hybrid_search, and gives a 3-5 line summary plus a link to open it in the launcher's DiagramsTab. Output may be incomplete, inaccurate, or visually broken — always tell the user to verify against the source code before sharing. Use when the user asks "show me a call graph for X", "what does Y depend on", "draw the inheritance chain rooted at Z", or any other request that maps to a small-to-medium subgraph extraction from the code graph.

Coordinates multi-channel incident response for a consulting engagement - drafts war-room updates, status-page entries, client comms, and the post-incident review; use when a client production issue is active or just resolved

Builds a 30/60/90-day cross-platform content calendar for a vendor across Instagram, TikTok, LinkedIn, X, YouTube Shorts, and newsletter. Takes a theme and cadence as input, produces a dated calendar with hooks, hashtags, CTAs, and a repurposing graph showing which posts feed which platforms. Use when the user says "plan content for next month/quarter", "build a content calendar", or "I need 30 days of posts".

Guide for using /compact (Claude Code built-in) with the pre-compact save pipeline. Shows what gets saved before compression and what gets reinjected after.

Expert guidance on database design, schema optimization, query performance, and database technology selection

Investigate complex bugs, intermittent failures, performance degradations, and system-wide issues requiring deep reasoning across multiple components

Deployment strategy guidance - platform selection, CI/CD pipeline design, environment configuration, monitoring

Quick GUI/UX/UI design consultations and recommendations

Quick VRAM/RAM calculations, hardware recommendations, feasibility checks for AI models

Designs the idempotency strategy for a state-changing operation - key derivation, storage choice, TTL, collision handling, and threading through downstream side effects. Use when adding a new endpoint with side effects, hardening an existing one, or designing a workflow that survives retries

Writes Adobe Photoshop automation scripts in UXP (JavaScript) or legacy ExtendScript (JSX), and GIMP scripts in Python (Script-Fu fallback). Use when a repetitive Photoshop task needs to run on dozens of files, when a custom panel/plugin is wanted, or when an Action recorder won't capture the logic needed.

Audits a scientific project for reproducibility — environment pinning, seed setting, data hashing, notebook discipline, workflow orchestration, and provenance capture. Use when the user asks "is this project reproducible", "what's missing for a reviewer to rerun this", "audit my repo before submission", "why do I get different results each run", or before submitting code with a manuscript.

Performs a diagnostic on a SaaS product's revenue and retention metrics from a CSV or spreadsheet of customers/subscriptions/events. Computes MRR, ARR, churn (gross/net), LTV, CAC, payback period, quick ratio, and cohort retention; flags anomalies; identifies the single highest-leverage fix. Invoked when the user asks "are my SaaS metrics healthy", "compute LTV for my data", "is my churn high", "review my MRR breakdown", or shares a billing/customer CSV.

Cross-layer security analysis (frontend XSS/CSRF, backend injection, AI prompt injection, infrastructure)

Designs SLIs, SLOs, and multi-window multi-burn-rate alerts from a service description, then emits the Prometheus recording rules and alerting rules. Invoke when a service needs its first SLO, when an existing threshold-alert is flapping or missing real incidents, or when an SRE team is operationalising error budgets.

Recommends the appropriate statistical test given a data description and research question, including assumption checks, alternatives if violated, sample-size guidance, and effect-size reporting. Use when the user asks "what test should I use", "is this t-test the right choice", "how many subjects do I need", "what's the right way to analyse this dataset", or describes a dataset + hypothesis without a chosen method.

Reviews Terraform/OpenTofu plan output for destructive changes, drift, IAM expansions, hardcoded values, and unsafe resource recreations before apply. Invoke when the user shares plan output, when a CI plan job posts a diff to a PR, or before any non-trivial production apply.

React best practices - component patterns, state management selection, performance optimization, testing strategies

Analyses a SaaS product's pricing page (current and competitor) and produces a concrete redesign with tier structure, value metric, anchoring, annual-discount math, and a measurable rollout plan. Invoked when the user asks "is my pricing right", "review my pricing page", "compare my pricing to [competitor]", "how should I raise prices", "should I add a third tier", or shares a competitor pricing URL with intent to react.

Produces a one-pager prep doc for an upcoming sales call (discovery, demo, negotiation). Given a prospect's LinkedIn URL and company URL (plus optionally their last email/reply), generates company context, likely buying-committee map, expected pain points, discovery questions, objection-handling cheat sheet, and the first 60-second pitch. Use before any call where you have < 15 min to prep.

Generates a complete SEO content brief from a target keyword and SERP context. Includes search intent, AI Overview likelihood, outline with H2/H3 questions, internal/external linking, on-page metadata (title/meta/schema), and a 2026-aware angle. Use when the user says "write me a brief for keyword X", "I'm targeting [keyword]", or "outline a post about [topic]" and they want SEO traffic.

Builds a reliable LLM-powered extraction pipeline for messy inputs (PDFs, emails, transcripts, HTML) into a strict JSON schema with validation, automated correction loop, and observability. Use when designing extraction from unstructured documents or hardening one that fails too often

Break complex features into implementable tasks with estimates, dependencies, and risk assessment

Hardens an inbound webhook endpoint with HMAC signature verification, timestamp replay protection, idempotency, schema validation, async processing, and dead-letter handling. Use when designing a new webhook receiver or auditing an existing one

Calculates the realistic monthly cost envelope for a workflow design including LLM tokens, third-party API spend, compute/queue infrastructure, and per-vendor pricing tiers. Use before committing to an engine or before a workflow ships to production