/humanize-rlcr

name: humanize-rlcr
description: Start RLCR (Ralph-Loop with Codex Review) on Codex using the native Stop hook.
type: flow
user-invocable: false
disable-model-invocation: true

Humanize RLCR Loop

Use this flow as the Codex entrypoint for RLCR. Codex installs of Humanize require native hooks support and install the Humanize `Stop` hooks automatically.

Runtime Root

The installer hydrates this skill with an absolute runtime root path:

{{HUMANIZE_RUNTIME_ROOT}}

All commands below assume `{{HUMANIZE_RUNTIME_ROOT}}`.

Required Sequence

1. Setup

Start the loop with the setup script:

"{{HUMANIZE_RUNTIME_ROOT}}/scripts/setup-rlcr-loop.sh" $ARGUMENTS

If setup exits non-zero, stop and report the error.

2. Work Round

For each round:

1. Read current loop prompt from `.humanize/rlcr/<timestamp>/round-<N>-prompt.md` (or `finalize` prompt files when in finalize phase). 2. Implement required changes. 3. Commit changes. 4. Write required summary file:

• Normal phase: `.humanize/rlcr/<timestamp>/round-<N>-summary.md`
• Finalize phase: `.humanize/rlcr/<timestamp>/finalize-summary.md`

5. Stop or exit normally. 6. Let the native Humanize `Stop` hook run automatically. 7. If the hook blocks exit, follow the returned instructions exactly and continue the next round.

What This Enforces

The native Stop-hook path enforces:

• state/schema validation (`current_round`, `max_iterations`, `review_started`, `base_branch`, etc.)
• branch consistency checks
• plan-file integrity checks (when applicable)
• incomplete Task/Todo blocking
• git-clean requirement before exit
• `--push-every-round` unpushed-commit blocking
• summary presence checks
• max-iteration handling
• full-alignment rounds (`--full-review-round`)
• strict `COMPLETE`/`STOP` marker handling
• review-phase transition guard (`.review-phase-started` marker)
• code-review gating on `[P0-9]` markers
• hard blocking on codex review failure or empty output
• open-question handling when `ask_codex_question=true`

Critical Rules

1. Never manually edit `state.md` or `finalize-state.md`. 2. Never skip a blocked hook result by declaring completion manually. 3. Never run ad-hoc `codex exec` / `codex review` in place of the hook-managed phase transitions. 4. Always use files generated by the loop (`round-*-prompt.md`, `round-*-review-result.md`) as source of truth.

Options

Pass these through `setup-rlcr-loop.sh`:

| Option | Description | Default | |--------|-------------|---------| | `path/to/plan.md` | Plan file path | Required unless `--skip-impl` | | `--plan-file <path>` | Explicit plan path | - | | `--track-plan-file` | Enforce tracked plan immutability | false | | `--max N` | Maximum iterations | 42 | | `--codex-model MODEL:EFFORT` | Codex model and effort for `codex exec` | gpt-5.5:high | | `--codex-timeout SECONDS` | Codex timeout | 5400 | | `--base-branch BRANCH` | Base for review phase | auto-detect | | `--full-review-round N` | Full alignment interval | 5 | | `--skip-impl` | Start directly in review path | false | | `--push-every-round` | Require push each round | false | | `--claude-answer-codex` | Let Claude answer open questions directly | false | | `--agent-teams` | Enable agent teams mode | false | | `--yolo` | Skip quiz and enable --claude-answer-codex | false | | `--skip-quiz` | Skip Plan Understanding Quiz (implicit in skill mode) | false |

Review phase `codex review` runs with `gpt-5.5:high`.

Usage

# Start with plan file
/flow:humanize-rlcr path/to/plan.md

# Review-only mode
/flow:humanize-rlcr --skip-impl

Cancel

"{{HUMANIZE_RUNTIME_ROOT}}/scripts/cancel-rlcr-loop.sh"