craft-squad

Craft (squad variant: Claude + Codex + local LLM)

Full design-first pipeline with every AI component active. Combines /craft-duo (Codex for data layer) with /craft-local (LM Studio for review/exploration). Optionally enables a skeptical adversarial review from Codex via codex-plugin-cc if installed.

Profile: claude+codex+llm

Pipeline: Brainstorm → Plan → Develop (hybrid executors + LLM reviews + optional adversarial review) → Browser Test → Report

Requirements

• Codex CLI installed: npm i -g @openai/codex
• Codex authenticated: codex login or OPENAI_API_KEY
• LM Studio running at ${LLM_URL:-http://127.0.0.1:1234}
• AGENTS.md at project root (auto-generated)
• Optional: codex-plugin-cc installed in Claude Code — enables adversarial review bonus

How this wrapper works

Writes the profile marker, then delegates to the canonical craft pipeline.

Step 1: Write profile marker

echo -n "claude+codex+llm" > .craft-profile

Step 2: Follow the craft pipeline

Read skills/craft/SKILL.md and follow every phase. All profile gates are satisfied: LLM steps run, Codex routing activates, and the develop skill checks for codex-plugin-cc to optionally run an adversarial review.

The user input is: $ARGUMENTS

Expected behavior

• All LLM steps run (LM Studio required)
• All Codex routing runs (Codex CLI required)
• If codex-plugin-cc is installed, develop Step 3.5 invokes /codex:adversarial-review on the diff as a skeptical second-opinion review
• If codex-plugin-cc is NOT installed, the adversarial review step is silently skipped — no failure

When to use

Use /craft-squad when:

• Stakes are high and you want every layer of review
• The feature is complex enough that a skeptical second opinion matters
• You have both LM Studio and Codex CLI available

For most day-to-day work, /craft-duo (no LLM) or /craft (no external deps) are usually enough.