Impact in one sentence
Deletes, installs, tests, git commands, file writes, and edits become short plain-English previews.
CLI
Explain agent actions. Before they run.
NotATechBro shows a plain-English preview whenever Claude Code, Codex, or Hermes asks to run a command.
Know what the agent is asking for.
When an agent asks for permission, NotATechBro explains the command before you choose yes or no.
3
Agent CLI families: Claude Code, Codex-style payloads, and Hermes.
0
Cloud calls, telemetry, accounts, or mandatory LLMs at runtime.
22
Tests covering shell rules, file rules, adapters, CLI output, and malformed JSON.
Product
Built for the moment before “approve”.
New agentic coding users do not need a shell lesson. They need to know what the agent is about to do.
Works across agent hooks
Small adapters accept common Claude Code, Hermes, and Codex-style payloads. The rules stay local and easy to inspect.
Explanations, not false confidence
NotATechBro explains only. It does not claim a command is safe, and it keeps hook output separate from human text.
Install
Start with one local install.
Install the CLI, run a sample payload, then wire it into your agent's pre-tool hook.
Learn more
Docs for setting it up without guessing.
Quickstart
Install the CLI and run the first preview payload.
Read quickstart →CLI adapters
Claude Code, Hermes, Codex-style payloads, and generic tool JSON.
View adapters →Safety wording
Explain risk without pretending to approve commands.
Review language →Architecture
Hook payload, normalizer, action model, rule engine, preview.
See architecture →FAQ
Questions before you wire it into your agent.
Does it block dangerous commands?
No. This version explains what is likely to happen. It does not approve, deny, block, or change tool calls.
Does it send commands or paths to a server?
No. It runs locally and uses deterministic rules. There are no runtime LLM calls, cloud calls, accounts, or telemetry.
What will I see before a command runs?
You see the command type, the command itself, and a short explanation of what it is likely to do. Then your agent still asks you to choose yes or no.
Who is this for?
It is for people using agentic coding tools who want plain English before they approve a command, especially if shell syntax feels unfamiliar.
Is Codex fully verified?
Codex-style payloads are supported, but the exact hook config should be checked against the Codex CLI version you use.
Try NotATechBro.
Give new agentic coding users a clearer moment before the agent runs a command.