riel-codex-deepdesign

name	riel-codex-deepdesign
description	Use when running, editing, or troubleshooting the packaged codex-deepdesign rielflow workflow for design-doc specification authoring with deep and broad review loops.

Riel Codex Deepdesign

Use this skill when the user asks Codex to create or refine design-doc specifications through the packaged codex-deepdesign workflow, or when maintaining that workflow package.

Workflow Bundle

• Workflow id: codex-deepdesign
• Package id: codex-deepdesign
• Backend: codex-agent
• Primary output: accepted design-docs/ specification plus deep, broad, and adversarial review summaries
• Loop rule: Node 2, Node 3, and Node 4 route back to Node 1 until no review reports high or middle findings

Standard Run

Run from the repository root after checkout or installation:

rielflow workflow run codex-deepdesign \
  --variables '{"workflowInput":{"feature":"Describe the feature to design.","targetDesignDoc":"design-docs/specs/design-example.md","constraints":["Design documentation only."],"acceptanceCriteria":["Deep review has no high or middle findings.","Broad review has no high or middle findings.","Adversarial review has no high or middle findings."]}}' \
  --output json --no-auto-improve

Adjust feature, targetDesignDoc, targetScope, constraints, acceptanceCriteria, and nonGoals to the user's request.

Provisional Decisions

The workflow should not stop for user confirmation. When a user decision would normally be required, Node 1 must choose the generally preferable, conservative specification and document it as a provisional decision.

Each provisional decision should include:

• the chosen provisional specification
• why user confirmation would normally be needed
• why this provisional choice is preferable
• what should change if the user later disagrees

Prefer recording these decisions in a clearly named Provisional Decisions section of the target design document. Use design-docs/user-qa/ when the decision needs a separate follow-up note.

Review Expectations

Node 2 deep review checks assumptions, prerequisites, irregular states, lifecycle edge cases, recovery paths, malformed inputs, permissions, privacy, observability, and testability.

Node 3 broad review checks existing similar features, duplication risk, adjacent workflow behavior, package and installation impacts, combined use cases, compatibility, migration, docs, and repository conventions.

Node 4 adversarial review checks failure modes, misuse paths, misleading success states, implementation loopholes, overconfident assumptions, and verification that could pass while material behavior remains broken.

All reviewers must return adapter JSON with when.needs_revision and payload.needs_revision. Set those booleans to true only when any high or middle finding exists.

Verification

Validate the package workflow after edits:

rielflow workflow validate codex-deepdesign \
  --workflow-definition-dir ./packages/codex-deepdesign/workflows

Run the bundled mock scenario when prompts, transitions, or review payloads change:

rielflow workflow run codex-deepdesign \
  --workflow-definition-dir ./packages/codex-deepdesign/workflows \
  --mock-scenario ./packages/codex-deepdesign/workflows/codex-deepdesign/mock-scenario.json \
  --output json

Refresh rielflow-package.json digests after changing the workflow or packaged skill:

bun .agents/skills/rielflow-package-release/scripts/update-package-digests.ts codex-deepdesign