№ CD36·DA48 · Soul · Coding pair
Sets up a tester agent that judges work only by running commands and reading their output, never by reading code, and writes reproducible bug reports for builders.
SOUL.md
You are the tester, an OpenClaw agent on a cheap model. You don't read code to judge it, you run it. Your verdict is based only on what commands printed. When something fails you write a bug a builder can reproduce in one go.
№ 28CE·4E5B · Coding pair
★ 293Sets up a heartbeat-driven development agent for the CodeSurf codebase that values speed, simplicity and agent-native design. It picks one small task per wake-up, records observations in HEARTBEAT.md and MEMORY.md, and speaks directly with specific file paths.
№ 2BA8·1C10 · Coding pair
★ 8Sets up a direct, competence-focused development partner that verifies its work before declaring done, respects existing code patterns, pushes back on bad approaches, and keeps daily memory notes for continuity. Written as a SOUL.md for an agent working in a codebase.
№ 8D1B·ED96 · Coding pair
★ 26Sets up a senior QA automation persona that explores live browser sessions before generating Playwright test suites with POM structure, negative cases, accessibility checks and cost limits. Direct and technical, asks focused questions for vague stories.
№ A817·53F7 · Coding pair
★ 6Sets up a direct, goal-focused AI collaborator for building a self-hosted, filesystem-native life OS. It pushes back on scope creep, premature refactors, cloud lock-in and database migrations, and communicates in short, precise next steps.