cupnfish/humanize-rlcr

Humanize RLCR Loop (Hook-Equivalent)

Use this flow to run RLCR in environments without native hooks.
Do not re-implement review logic manually. Always call the RLCR stop gate wrapper through the CLI:

humanize gate rlcr

The gate command routes through the same stop-hook-compatible logic, so skill-mode behavior stays aligned with hook-mode behavior.

Runtime Command

All commands below assume the humanize CLI is available on PATH.

Required Sequence

1. Setup

Start the loop with the setup command:

humanize setup rlcr $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 the required summary file:
   • normal phase: .humanize/rlcr/<timestamp>/round-<N>-summary.md
   • finalize phase: .humanize/rlcr/<timestamp>/finalize-summary.md
5. Run the gate command:

humanize gate rlcr
humanize gate rlcr --session-id "$CLAUDE_SESSION_ID"
humanize gate rlcr --transcript-path "$CLAUDE_TRANSCRIPT_PATH"

6. Handle gate result:
   • 0: loop is allowed to exit
   • 10: blocked by RLCR logic; follow returned instructions exactly and continue the next round
   • 20: infrastructure/runtime error; report the error and stop

What This Enforces

By routing through the native gate logic, this skill 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 guards
• 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 gate result by declaring completion manually.
3. Never run ad-hoc codex exec or codex review in place of the gate for phase transitions.
4. Always use loop-generated files (round-*-prompt.md, round-*-review-result.md) as source of truth.

Options

Pass these through humanize setup rlcr:

| 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 or --max-iterations N | Maximum iterations | 42 | | --codex-model MODEL:EFFORT | Codex model and effort for codex exec | gpt-5.4:xhigh | | --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 |

Review phase codex review runs with gpt-5.4:high.

Usage

# Start with plan file
humanize setup rlcr path/to/plan.md

# Review-only mode
humanize setup rlcr --skip-impl

Cancel

humanize cancel rlcr