GKjohns

Push heavy work into sub-agents and keep the main conversation's context clean. Use when starting any multi-step task — codebase research, sprint execution, browser/Playwright verification, or anything that would otherwise dump large output (screenshots, file listings, wide search results) into the main loop. Bring back only the conclusion.

Turn a narrated video walkthrough of a space into a room-by-room cleanup and redecoration plan — a markdown checklist plus a shareable, styled HTML one-pager with a budget-tuned shopping list. Use when someone records themselves walking through a home, office, or room narrating what they want thrown out, moved, cleaned, organized, or bought, and wants an organized plan to execute. Triggers include "here's a walkthrough video of my place", "turn this video into a cleanup plan", "spring cleaning video", "process this room tour", or "what needs to happen to get this room from its current state to what I want".

Make sense of large PDF dumps — one massive PDF or many — by building a cached, searchable workspace with OCR, per-page index, section splits, and a standalone summary. Use when a user hands over hundreds or thousands of pages and wants to know what's in there, search across it, or hand a digest to someone else.

Design a logo for a Kyle Johnson / Monument Labs project. Use whenever the user asks for a logo, logomark, wordmark, brand mark, app icon, or favicon — or when they ask you to "design a logo for X" or "make a mark for X". Produces an HTML gallery of at least five SVG concepts in Kyle's house style (simple, clever, semantically tied to the name, renderable in pure SVG, theme-friendly), lets the user pick one, then delivers the chosen logo as a clean standalone SVG.

Autonomous data investigation agent. Invoke when a user has a dataset (CSV, Parquet, or DuckDB) and asks an analytical question requiring multiple queries — metric changes, anomaly detection, root cause analysis, segmentation, or pattern discovery. Runs SQL against a local DuckDB database, builds an evidence graph, validates hypotheses, and produces an auditable brief.

The real operator know-how for getting a good edit out of the Agentic Image Editor's toolset (straighten, crop, exposure, tone, toneCurve, whiteBalance, contrast, vibrance, saturation, splitTone, dehaze, denoise, look, sharpen). The taste and judgment a working editor has internalized: how to read an image, genre-aware sensibility, the opinionated-but-collaborative stance, order of operations, target values, intent/mood translation, and the failure modes. Read this before tuning the agent's prompt or reasoning about why an edit went sideways. Keep it in lockstep with `src/server/utils/editing-guide.ts` (the token-lean version injected at runtime).

Install and deploy an OpenClaw gateway on Google Cloud Platform from scratch. Walks through GCP project setup, VM creation, Docker installation, OpenClaw deployment, SSH tunnel configuration, and verification. Use when setting up OpenClaw on GCP, creating a VM for an OpenClaw gateway, deploying OpenClaw to the cloud, or connecting to a remote OpenClaw instance.