rodrigoterra

agents-discipline

Agents Discipline & Good Project Management for any Claude Code project. Bootstraps the 'quiet files' (CLAUDE.md, AGENTS.md, INDEX.md, specs/, .claude/hooks, rules, output-styles, settings) and wires AGENTS.md so OpenAI Codex can adversarially review what Claude shipped.

Annotate screenshots with shapes, arrows, numbered markers, redaction boxes and per-mark text notes in a local browser canvas, then hand the flattened image plus a compact JSON sidecar straight back into Claude's context. Supply images by path, paste, drag-drop, or picker. Built on a portable TypeScript MCP server (reusable by Codex and other MCP clients).

Behavioral guidelines based on Andrej Karpathy's observations of LLM coding mistakes. Helps Claude resist overcomplication, make surgical changes, surface assumptions, avoid defensive bloat, and define verifiable success criteria. Vendored from multica-ai/andrej-karpathy-skills (MIT).