oh-my-humanize

The terminal coding agent we built first: hash-anchored edits, an optimised tool harness, LSP, Python, browser control and subagents, all in one native binary — built for long-running agentised workflows before there was a runtime to run them under.

humanfia/oh-my-humanize · humanfia/humanize-plugin

What it is

A workflow-native terminal agent. Where most terminal agents are a chat loop with tools bolted on, this one was built around the assumption that a session runs for hours: edits are anchored by content hash so a stale line number cannot corrupt a file, the tool harness is tuned for the calls agents actually make, and code intelligence, debugging, browser control and subagents are in the harness rather than in a plugin.

humanize-plugin is the same idea packaged for the coding agent you already use, if you would rather keep your CLI and take the workflow.

Where it sits now

oh-my-humanize is the origin of the position everything here is built on: that the leverage in agentic development is in the workflow, not the wrapper. Building it is how we learned that the workflow deserved to be a first-class object — a thing you can read, fork, publish and score — rather than behaviour compiled into one binary.

That conclusion is Humanize 2, which drives whichever coding agent you already have instead of replacing it, and flowverse, where the workflows live as code.

If you want a single native harness and are happy inside it, oh-my-humanize is there and it is Apache-2.0. If you want your existing CLI driven by a loop you can change, start with Humanize 2.