مهارات جديدة

Arquitetar projetos, processos e sistemas completos no método Sínkra: reconstruir AS-IS, modelar entidades, lifecycles, stages, transições e relações, resolver decisões, desenhar TO-BE, contratar Humanos, Workers e Agentes, forjar skills orquestradoras e publicar um sistema auditável. Usar quando o pedido envolver desenho operacional, arquitetura de processos, automação ou estrutura de agentes conectada ao sistema Sínkra.

Deliver a ready issue end to end, unattended — a task, a bug, or a whole feature:ready feature (split it into the tasks its Technical spec lists; each task is then delivered by its own run as it turns ready). Claims, delegates implementation, the two blind reviews and the simulator check (in parallel) to fresh subagents, reasons only over their report blocks, and merges each PR through the gate so the merge ships an OTA update. Use when someone says "deliver

Use when a page must paint before its JavaScript runs.

Dentifies specific cats by comparing images or videos of their faces. It supports distinguishing between different individuals in multi-cat households, ensuring accurate recognition once the data is registered in the database. | 猫脸识别技能，通过猫脸图片/视频比对识别具体是哪只猫咪，支持多猫家庭区分不同个体，录入底库后可准确识别

Assemble a full company due-diligence report — entity resolution, registry profile, ownership and related parties, business and industry-chain position, risk records, adverse media, and (for issuers) financial red flags. Triggers on "尽调", "尽职调查", "due diligence", "DD报告", "排查这家公司", "供应商准入", "交易对手调查".

Build, fix, or check a CloudTalk AI Voice Agent: its v2 "Expert Mode" configuration. Connected to CloudTalk, it creates, reads and saves the agent directly against your real account (numbers, knowledge bases, tools, transfer targets; every save needs your approval, and a change to an existing agent can be undone); offline, it returns the JSON you paste into the dashboard's Advanced/code tab. Use whenever the user wants to create or edit ANY inbound or outbound AI voice/phone agent for CloudTalk, for any use case or industry (reception, qualification, support, booking, surveys, reminders, notifications, lead capture and more; these are just examples, not the limit), or asks for an AI Voice Agent config, a goalPrompt/greeting/scenarios/skills, or pastes an existing config to review, edit, or reuse. Produces a behaviorally-sound, save-ready config that reuses the agent's existing setup, and runs a structural + behavioral validator first.

Turns the approved transition choices and scene order into typed operations. Prefers plain cuts; uses fades or wipes only where a reviewed recipe and the approved plan call for them. Use when assembling scenes.

Use when a Python or TypeScript app sends large tool outputs, logs, JSON, search results or long chat history to an LLM and the task is to cut input tokens, reduce token cost, fit a context window, or trim tool results without losing errors. Adds the headroom-ai compress() call; in-process, no API key, no model download.

Analyze cross-channel marketing performance from live Coupler.io data through its hosted MCP server: channel and campaign comparison, ROI, anomalies and budget insights.

Plan the "Comment KEYWORD" CTA of an Instagram Reel (keyword, overlay, caption, DM copy), then, after explicit user confirmation, set up the matching auto-DM in CommentKeyword.

Stress-test a component with realistic worst-case data through a development-only fixture toggle and report verified failures.

Use Bowmark when a task depends on what a live website shows right now: current prices, availability or stock, fares and quotes, a site search, or data behind a form or a login. The agent writes a short JavaScript script against typed site functions and gets typed results back.