Pack
AI / agent engineer (LLM apps, evals, MCP)
AI coding-agent instructions for: Building LLM-powered features and agents in TypeScript: a swappable model interface with deterministic fakes, budgeted agent loops, human…
Starter agents are role definitions you can drop into your agent tool: a clear job, rules, suggested routines, the access it needs, safety limits and test scenarios. Start from a working role instead of a blank prompt.
Pack
AI coding-agent instructions for: Building LLM-powered features and agents in TypeScript: a swappable model interface with deterministic fakes, budgeted agent loops, human…
Pack
A briefing agent that collects what matters for one person's projects every weekday morning and delivers one short, scannable message — silent when nothing changed.
Pack
A documentation agent that reads your code and writes accurate, tested docs organised by type (tutorial, how-to, reference, explanation).
Pack
A design assistant for indie developers: turns ideas into one-page GDDs, defines core loops, systems and balance tables, and checks numbers with small simulations.
Pack
A marketing assistant for solo developers: positions a product, plans small launches and drafts posts and store text — everything stays a draft until you approve it.
Pack
A review agent that reads a diff, checks it against the repo's rules and tests, and returns a short prioritised review — it never approves, merges or pushes.
Pack
A testing agent that turns a feature or bug report into a test plan, runs it, and returns reproducible findings — it reports bugs, it doesn't hide them.
Pack
A release agent that prepares versions, changelogs and checklists and tracks readiness — and never tags, publishes or deploys without an explicit go.
Pack
A research agent that answers one question at a time with sources, confidence levels and clearly marked unknowns — no vibes, no invented citations.
Pack
A support agent that sorts incoming requests, finds known answers, drafts replies and escalates what needs a human — nothing is sent without approval.
Pack
A coding agent that does one job: turn a small, clear task into a verified change using red-green-refactor, and report only with evidence.