
Claude Skills by CarlosCaPe
github.com/CarlosCaPeMarco de investigacion de mercado a nivel consultoria, 12 modulos del sizing a la estrategia de entrada. Dispara con 'market research', 'TAM/SAM/SOM', 'analisis competitivo', 'segmentacion de clientes', 'SWOT', 'Porter', 'pricing', 'go-to-market'.
Instala y configura el stack MCP de un arm respetando Arm Isolation: los MCP de nube van a scope user, los de base de datos a .mcp.json dentro del repo del cliente. Dispara con 'setup MCP para X', 'agregar MCP al arm', o al dar de alta un arm nuevo.
Enruta el trabajo por complejidad a traves del ladder de modelos (mechanical, bulk, build, judgment), vendor-agnostico. Carga al abanicar sub-agentes o al elegir el override de model en un Workflow/Agent.
Multi-Object Rename Convergence
Audit a long technical document for hallucinations and uncited claims by dispatching N specialist agents in parallel, each scoped to a domain. Each returns a severity-tiered table; the orchestrator consolidates and decides surgical-fix vs full-rebuild.
Loop diario sobre feeds RSS de IA (Willison, OpenAI, DeepMind, arXiv cs.AI): filtra los articulos que ensenan algo que el cerebro aun no sabe. Promueve un keeper a skill real con /news-promote.
Los documentos llevan solo contenido vigente; git es el control de versiones. Nunca tachado para marcar lo hecho, nunca texto viejo junto al nuevo 'para historia', nunca secciones comentadas 'de referencia'. Aplica a toda edicion o revision de documento.
Cuando un mismo valor vale 10 en un lado y 12 en otro, converge ambos a un canonico referenciado en todas partes, nunca bifurcado. Dispara al editar tokens de diseno, estilos o convenciones en varios archivos, y ante 'converge' o 'armoniza'.
Los tres anclajes de Octorato (pulpo, Linux, teseracto) son una misma estructura mostrada tres veces, no tres marcas pegadas. Nombra el metodo de abstraccion. Invoca al explicar por que cohesionan o al anadir un anclaje nuevo.
The two symbolic anchors of the Octorato framework — the 8 → ∞ (lay the octopus's eight arms sideways and you get the lemniscate, the unbounded) and the Tesseract → 4D (the four-phase paradigm is the 4-dimensional analog of a cube). Invoke when explaining naming rationale, brand identity, the philosophical layer, or when authoring launch comms / talks / podcasts that touch the framework's depth.
OpenStock is an open-source alternative to expensive market platforms. Track real-time prices, set personalized alerts, and explore detailed company insights — built openly, for ev
Create and execute phased implementation plans with documentation discovery and subagent delegation. Use when asked to plan a feature, execute a multi-step task, or when 2D Delegate determines the task needs structured planning before execution.
Orphan Data Detection & FK Rollout
Determine code/file ownership by git authorship (log/blame), never by which repo it lives in. Load before assigning work, escalating a fix to "another team", or routing a problem by repo location — the author may be on your own team. Repo location is not ownership.
When a Cloudflare Pages Function (or any Worker-style serverless handler) returns a generic 502/500 hiding the actual error, instrument the handler with a `?debug=1` query param + an in-memory checkpoint array that accumulates step markers and gets returned in the response body. Lets you bisect "which step crashed" in 1-2 deploy/retest cycles instead of guessing for hours.
Peer Review Lifecycle for Technical Documents
Decision rule for WHEN to fan work out to N parallel sub-agents/arms vs solve it centrally, and HOW the center reconciles divergent results without becoming the bottleneck. Use before launching multiple agents or a Workflow, or when one context can't hold the whole job.
pg_cron Scheduled Maintenance
pg_stat_statements Query Observability
Multi-stage RAG ingestion pipeline for regulated content. Fetch → text-normalize → PHI-screen → chunk → embed → route → store → digest. Default-deny routing (local LLM for tainted, cloud LLM only for allowlisted clean sources). Per-user folder isolation via sha256(email)[:12]. Manual slash-command trigger principle for sensitive workloads.
Post-Check Verification
On repos where pushing to main auto-deploys to production, default to PR-based workflow even for self-authored changes — PR adds the safety pass that a direct push skips
Before arming auto-merge on a PR that touches production code, dispatch a QA specialist agent against an explicit test/user-case spec for the diff. Build-passes ≠ correct. Auto-merge is the merge mechanism, not the gate.
Primary Key Coverage & Identity Columns
Procedure Rebuild via pg_get_functiondef
Production Bug Fix in Stored Functions
Token-optimized code exploration using index-first, fetch-on-demand principle. Reduces token consumption 4-8x vs reading full files. Use when exploring large codebases, navigating files over 100 lines, or when token budget matters.
Generate a "Journey Into [Project]" narrative report analyzing a project's entire development history from git log, CHANGELOG, and release notes. Use when user asks for a timeline report, project history analysis, development journey, full project report, or "what's the story of this project?". No external dependencies.
Generates optimized prompts for AI tools. Activates only when the user explicitly asks to write, fix, improve, or adapt a prompt for a specific AI tool (LLM, Cursor, Midjourney, image AI, video AI, coding agents, etc.). Does not activate for general conversation, coding tasks, document writing, or other non-prompt-engineering work.
QA Deployment Prompts
Skill hijo de querymaster para Azure Data Explorer. ADX usa KQL, NUNCA SQL. Conexion azure-kusto-data con SSO y buenas practicas para telemetria en tiempo real. Se activa cuando el motor resuelto es ADX.
Skill hijo de querymaster para Databricks. PLACEHOLDER: aun no hay workspace ni cluster provisionado, solo la plantilla de conexion con databricks-sql-connector. Se completa cuando exista acceso.
Skill hijo de querymaster para PostgreSQL (estandar, Azure Database, Aurora): patrones de conexion psycopg2 y Node, sesion readonly y buenas practicas de consulta. Se activa cuando el motor resuelto es PostgreSQL.
Skill hijo de querymaster para Snowflake: conexion por Browser SSO, OAuth y usuario/contrasena de servicio, REST SQL API y buenas practicas de warehouse. Se activa cuando el motor resuelto es Snowflake.
Skill hijo de querymaster para SQLite: bases locales en archivo, storage de Optuna, barridos analiticos y uso embebido, con PRAGMA de WAL y foreign_keys. Se activa cuando el motor resuelto es SQLite.
Skill hijo de querymaster para SQL Server y Azure SQL (Database, Managed Instance, on-prem): conexion pyodbc con token de Azure AD y buenas practicas T-SQL. Se activa cuando el motor resuelto es SQL Server.
Router maestro multi-motor: recibe un prompt en lenguaje natural, identifica el motor (PostgreSQL, Snowflake, SQL Server, ADX, SQLite, Databricks), genera la consulta y la ejecuta con dry-run por defecto. Dispara al pedir consultar una base, correr SQL o inspeccionar un esquema.
Range Partitioning for Growth Tables
When a rule or gate is chronically ignored, when designing enforcement for a new policy, or when the operator says a rule keeps failing. Teaches how to convert a passive "mental check" into a mechanical hook (ganglion) that fires without the central brain invoking it.
Analisis profundo de UN repo de GitHub: inventaria el codigo, extrae patrones que podrian volverse skill y escribe reporte en ~/.claude/knowledge/repo-deep-learn/. Usalo cuando el operador te entrega un repo y pregunta que se puede aprender de el.
Monitor diario de una watchlist de repos de GitHub: compara el HEAD SHA contra el snapshot previo y escribe un digest en ~/.claude/knowledge/repo-watch/. Usalo para seguir el pulso de competidores sin mirar GitHub todo el dia.
Empaqueta un repo completo, local o remoto, en un solo archivo optimizado para LLM, con compresion Tree-sitter (~70% menos tokens) y escaneo de secretos. Para auditar un repo entero o preparar contexto de un subagente.
Research Checklist Discipline
Decision rule for WHERE a new behavior or lesson belongs — the runtime layer (skill / memory / config / hook) vs the core source (CLAUDE.md). Default to the runtime layer; touch the core only when the change is a true invariant. Use whenever a lesson, preference, or new capability arrives and you must decide what to edit.
Build sandboxed applications for secure code execution. Load when building AI code execution, code interpreters, CI/CD systems, interactive dev environments, or executing untrusted code. Covers Sandbox SDK lifecycle, commands, files, code interpreter, and preview URLs. Biases towards retrieval from Cloudflare docs over pre-trained knowledge.
Rescata el transcript COMPLETO de una grabacion de reunion (Microsoft Stream, SharePoint) a partir de su URL, y entrega un .txt limpio, con procedencia y listo para compartir. Dispara con 'save-transcript', 'baja el transcript', 'saca el transcript de esta junta', 'transcript de <URL>'. Para audio crudo sin transcript ya renderizado usa transcribe.
Mantiene sincronizadas las tres capas de un proyecto schema-as-code: diagrama ER, diccionario de datos y el SQL declarativo. Define la cadena de verdad y el flujo de diff cuando divergen.
Get exact row counts per table in a given schema — PostgreSQL primary, with notes for other engines. Two-column result: table_name, row_count.
Schema Separation for ORM/Scaffolding Control
SDD step 7. Archive feature.md and plan.md into docs/specs-archive/<yyyymmddHHMM>-<feature-name>/ directory, then update docs/project.md with the new feature, and any architecture decisions made. Use after /sdd-review is complete and the feature is ready to merge.