A read-only Codex skill that pauses coding and explains code, systems, workflows, risks, and feature ideas in plain language with large ASCII diagrams when visuals help.
Machine Talk is a standalone Codex skill with one clear job: stop implementation mode, inspect the real project, and translate technical complexity into calm, human explanations for non-programmers, vibe-coders, founders, operators, and product-minded users.
- explains files, flows, APIs, architecture, and feature behavior in simple language
- draws large, readable ASCII diagrams for systems, pipelines, risks, and multi-step ideas when visualization helps
- translates jargon into normal human words
- stays strictly read-only and never edits code
- helps brainstorm product ideas, tradeoffs, and simplifications
- points out risks, hidden complexity, and side effects before you build
- vibe-coders and non-programmers who feel lost in code
- founders and operators who need technical clarity without implementation noise
- product-minded builders who want to discuss ideas before writing code
- Codex users who want a deliberate "don't code yet" mode
Ask Codex:
Use $skill-installer to install https://github.com/harnessmachine/machine-talk/tree/main/skills/machine-talk
You can hand that line to another Codex agent and it can install the skill by itself.
Manual install:
cp -R skills/machine-talk ~/.codex/skills/Restart Codex after installation.
Use $machine-talk to talk me through this codebase.Use $machine-talk to explain this feature like I'm not a programmer.Use $machine-talk to help me think through this before we code.Use $machine-talk to compare these options without editing anything.
Machine Talk uses ASCII diagrams when a topic has structure that should be seen, not only described:
- architecture
- request and data flows
- automations and agent workflows
- feature ideas with several moving parts
- risks, blockers, tradeoffs, and decision trees
The diagrams are written in the owner's language and include simple explanations inside the boxes, so the picture itself teaches how the thing works.
- skills/machine-talk - the installable skill
- README.md - English overview
- README.ru.md - Russian overview
- it explains before it suggests implementation
- it visualizes systems and pipelines with practical ASCII diagrams
- it keeps the tone human instead of programmer-to-programmer
- it stays critical and points at weak ideas, not just agreeable ones
- it never quietly slips into edit mode
MIT