Use when executing an implementation plan with Codex instead of Claude, after writing-plans produces a plan file
Send an implementation plan to Codex CLI for full-auto execution. Codex implements the entire plan, then you review the results.
superpowers:writing-plans produces a planexecuting-plans offers implementation choice and you choose CodexFind the most recent plan in docs/plans/ or ask the user which plan to execute.
Read ~/.claude/skills/multi-model-config.json to get the CLI command for the model assigned to implement.
CLAUDE.md (if exists) or project root configFill the template at codex-implement/prompt-template.md:
{PLAN_CONTENT} → full plan text{CONVENTIONS} → project conventions (from CLAUDE.md, .editorconfig, etc.)PROMPT_FILE=$(mktemp /tmp/codex-implement-XXXX.md)
# Write filled template to $PROMPT_FILE
# Read command from config
COMMAND=$(python3 -c "import json; c=json.load(open('$HOME/.claude/skills/multi-model-config.json')); print(c['cli'][c['models']['implement']]['command'])")
# Run implementation
$COMMAND $PROMPT_FILE
Note: The exact invocation flags depend on which CLI is configured. The user should update cli.codex.args and cli.codex.prompt_flag in the config to match their CLI's actual interface once installed.
After the external CLI finishes:
git log --oneline to see what was committedgit diff --stat to see what changedSay: "Codex implementation complete. Ready for multi-model review? Use /gemini-review and /codex-review."
Never:
superpowers:writing-plansgemini-review + codex-review (parallel)~/.claude/skills/multi-model-config.jsonmulti-model-orchestration skill manages the workflow choices.