official/humanize1
PolyArch/humanize — the Claude Code plugin humanize grew out of — as three flows, each set up on its own agents and stopping on its own. gen-idea opens a loose idea into a repo-grounded draft, gen-plan turns that draft into a plan both sides have converged on, and rlcr builds the plan under review until nothing is left to say.
hmz exec -f official/humanize1:gen-idea -a claude/claude-opus-5:max \
"add undo/redo to the editor"
hmz exec -f official/humanize1:gen-plan \
-a claude/claude-opus-5:max -a codex/gpt-5.6-sol:max "add undo/redo to the editor"
hmz exec -f official/humanize1:rlcr \
-a claude/claude-opus-5:max -a codex/gpt-5.6-sol:max "build it"Three rather than one because each is set up on its own: /agents asks one flow for the drafter, one for the planner and the analyst that reads it, and one for the builder and the reviewer. What passes between them is a file, as it is in the plugin — the draft, then the plan — so an idea may be opened on one model, planned on another and built on a third, with whatever reading and editing you like in between. They are three flows in one file.
Run it in a git repository: the work is anchored to the commit the plan was fixed in, and every review reads what came after it.
1 · gen-idea
One phase, one file, nothing kept. Running it again is meant to write another draft, which is why a run of it is never picked up.
One agent, n directions explored at once, one draft written out. --n and --output are the two settings, under the names the plugin gives them.
2 · gen-plan
The side that writes remembers and the side that reads does not — the rule the whole of humanize 1 is built on, here and again in the loop below.
The planner holds one session for the whole of the planning; the analyst arrives fresh each time and reads the plan against the repository. They converge, or the round limit stops them. --input names the draft to plan from, --discussion and --direct are the two modes, and alternative_plan_language writes a translated plan beside the plan.
3 · rlcr
The loop is a hook. A round ends when the builder believes the whole plan is done and tries to stop; what the reviewer says is what it hears instead of stopping.
The loop is a hook. The plugin blocks Claude's exit and puts the round to Codex there; so does this — a Moment.STOP hook on the builder, which is the same sentence. A round ends when the builder believes the whole plan is done and tries to stop, and what the reviewer says is what it hears instead of stopping. The plugin's tool validators are hooks too, on Moment.PERMISSION_REQUEST, which is why the builder has to be a backend that runs them.
Every flag the plugin takes is a field on that phase's own settings, under the plugin's own name for it: --max, --full-review-round, --skip-impl, --agent-teams, --yolo, --plan-file, --base-branch, and the rest. /config is all of them.
It writes what the plugin writes, where the plugin writes it: .humanize/rlcr/<timestamp>/ with state.md, goal-tracker.md, and a prompt, summary, contract and review per round — so humanize monitor rlcr reads a run of this.
Four things done another way
The plugin's mechanism, where humanize's is not the same mechanism:
codex review --base <ref> | Takes no prompt and is a Codex feature. Here the reviewer is whichever agent was chosen, so the code review is asked for, in a prompt that asks for exactly the [P0-9] output the loop then reads the same way. |
--codex-timeout | Cannot cut a turn short from here: a review that ran over is treated as a review that failed, which is the state the plugin's own timeout leaves the round in. |
/humanize:ask-codex | A task the plan tags analyze is a shell script the builder runs there. Here the builder has no way to reach the reviewer mid-round, so it is told to put the question in its round summary, where the reviewer answers it. |
| The plan quiz | Put to the person the way a coding agent's own question is put, so a run with nobody at the prompt is answered with nothing rather than waiting for a person who is not there. |
What it keeps
rlcr is a loop meant to run for days, so a run of it can be picked up: it keeps which.humanize/rlcr/ directory the loop is in and the round it reached, and reads state.md back as it stands rather than stamping a new directory beside a week of rounds. Everything else about the loop is already in that directory in the plugin's own format, and a second copy here would be a second place for it to be wrong.
A loop carries on with the settings it was set up with, which is what carrying on means. A run set up differently is neither quietly overridden nor quietly ignored: it says which setting it disagrees with the loop about, and starts a loop of its own. The agents are the one thing that is not a setting — /agents chooses them per run, and the state file is brought up to date to say who is reading the rounds.
gen-idea and gen-plan keep nothing. Each writes one file, running one again is meant to write another, and between their turns there is nothing a second run could honestly carry on from.
See also
- The moments of a turn — what a
STOPhook is, and the six others - official/rlar — the same actor-and-reviewer shape, without the plugin's format
- A flow that calls a flow — running one of these three from inside another flow