Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Start RLCR (Ralph-Loop with Codex Review) on Codex using the native Stop hook.
.claude/skills/humanize-rlcr/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-19 | ✗→✓ | ▲ Improved | — | — |
| case-04 | ✗→✓ | ▲ Improved | — | — |
| case-17 | ✗→✓ | ▲ Improved | — | — |
| case-20 | ✗→✓ | ▲ Improved | — | — |
| case-06 | ✗→✓ | ▲ Improved | — | — |
Use this flow as the Codex entrypoint for RLCR. Codex installs of Humanize require native hooks support and install the Humanize Stop hooks automatically.
The installer hydrates this skill with an absolute runtime root path:
bash{{HUMANIZE_RUNTIME_ROOT}}
All commands below assume {{HUMANIZE_RUNTIME_ROOT}}.
Start the loop with the setup script:
bash"{{HUMANIZE_RUNTIME_ROOT}}/scripts/setup-rlcr-loop.sh" $ARGUMENTS
If setup exits non-zero, stop and report the error.
For each round:
.humanize/rlcr/<timestamp>/round-<N>-prompt.md (or finalize prompt files when in finalize phase)..humanize/rlcr/<timestamp>/round-<N>-summary.md.humanize/rlcr/<timestamp>/finalize-summary.mdStop hook run automatically.The native Stop-hook path enforces:
current_round, max_iterations, review_started, base_branch, etc.)--push-every-round unpushed-commit blocking--full-review-round)COMPLETE/STOP marker handling.review-phase-started marker)[P0-9] markersask_codex_question=truestate.md or finalize-state.md.codex exec / codex review in place of the hook-managed phase transitions.round-*-prompt.md, round-*-review-result.md) as source of truth.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.
bash# Start with plan file /flow:humanize-rlcr path/to/plan.md # Review-only mode /flow:humanize-rlcr --skip-impl
bash"{{HUMANIZE_RUNTIME_ROOT}}/scripts/cancel-rlcr-loop.sh"
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-19 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-04 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-02 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-17 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-03 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-20 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-06 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-13 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-10 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-18 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-15 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-11 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-22 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-08 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-07 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-21 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-01 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-14 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-05 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-09 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-12 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-16 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 16 counted toward the lift figure. The other 6 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +50 percentage points is the difference between those two pass rates over the 16 comparable cases.
The per-case answers from this run were removed by the retention sweep, so the case table below shows the verdicts without the text either arm produced. The counts above were recorded at the time and are unaffected. Answers are now kept for 180 days.
Other measured skills in the registry, with their headline benchmark lift.