---
name: rdlc
description: Full RDLC workflow for Codex research projects: evidence gathering, confidence labels, audience-firewalled deliverables, fact regressions, slop scan, and review.
---

# RDLC

## Purpose

This is the Codex-native RDLC workflow skill. It is the honest equivalent of a missing `/rdlc` command.

Use it for research, fact-checking, drafting sourced prose, generated deliverables, data refreshes, and high-stakes review.

## Mandatory First Action

Before doing anything else, read the repo-local contract when present:

- `AGENTS.md`
- `RDLC.md`
- `README.md`
- `ROADMAP.md`
- `RDLC-LOOP.md`
- `START-RDLC.md`
- `PROVE-IT.md`
- `GOALS.md`

Repo-local guidance wins over generic skill guidance.

## Full Workflow

### 1. Plan

For meaningful work, make these visible before editing:

- claim or research question
- audience and deliverable
- source hierarchy
- confidence
- verification plan

Task routing gate: before giving execution steps, identify the execution lane as CLI, Desktop/computer-use, browser automation, or human-only setup. If browser sign-in, account pickers, MFA, admin portals, paywalls, or other stateful screens are involved, say `Desktop/computer-use` first, then provide handoff guardrails before CLI/browser steps.

Before implementation or drafting, do a docs update for the relevant workflow area when the change affects behavior, assumptions, or operator workflow.

### 2. Verify Sources

Prefer primary sources and authoritative records. Do not cite a source unless it supports the exact claim.

Use the configured vocabulary from `RDLC.md`:

- `VERIFIED`: multiple reliable sources or one primary source with cross-check
- `SUPPORTED`: one reliable source, no contradiction
- `INFERRED`: reasoning from verified premises
- `UNVERIFIED` or `GAP`: not yet proven or structurally unknowable

### 3. Draft Or Generate

Keep private, internal, and public audiences separated. The audience firewall belongs in generators and regressions, not just in prose comments.

### 4. Prove It

Run the relevant checks:

```bash
bash scripts/slop_scan.sh
bash scripts/regression_test.sh
python3 scripts/generate_deliverable.py
```

Every defect found by review becomes a permanent regression assertion.

### 5. Review

Self-review every claim, source, and confidence label. Use native Codex review for a second pass when the work warrants it:

- `codex review --uncommitted` before commit
- `codex review --base <branch>` for branch or PR-sized diffs
- `codex review --commit <sha>` for a specific commit

`review_model` controls native Codex review model selection. `auto_review` is for eligible approval prompts, not code-diff or research-diff review. Do not require `/autoreview` unless the current Codex host exposes it as a verified feature.

### 6. CI And Merge Guard

Never use auto-merge in this repo.

`NEVER AUTO-MERGE`

Read CI logs, handle valid review feedback, and merge explicitly only after the proof matches the diff.

### 7. Final Summary

Before handoff, make these visible:

- what changed
- what was verified
- what remains risky or unverified

### 8. Capture Learnings

If the session uncovered reusable lessons, capture them in the right local doc:

- `RDLC.md` for research/process rules
- `PROVE-IT.md` for proof gates
- `ARCHITECTURE.md` for generator or scaffold decisions
- `README.md` when the user-facing setup story changed

## Confidence Policy

- default: `xhigh`
- this repo uses `xhigh` as the normal path
- only drop lower when the user explicitly asks for it

## Hooks Vs Skill

- `skills = explicit workflow layer`
- `hooks = silent event enforcement`
- `repo docs = source of local truth`

## Feedback And Repo Focus

If the work is in a product repo, keep that session focused on the product repo. File a direct GitHub issue for proven reusable wizard findings and only switch to live wizard work if the product repo is actually blocked.

## Naming

- Canonical entrypoint: `$rdlc`
- do not pretend Codex has a native `/rdlc` command
- Do not create second public workflow names for the same contract.

## Learning Capture

When review exposes a reusable process defect, capture learnings in `RDLC.md` or the relevant repo doc before closing the slice.