From samocode
Runs OpenAI Codex CLI as a subagent for second opinions, code reviews, and questions. Useful when you want a different AI model's perspective.
How this skill is triggered — by the user, by Claude, or both
Slash command
/samocode:codexThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Run OpenAI Codex CLI (GPT-5.2) for second opinions and external reviews.
Run OpenAI Codex CLI (GPT-5.2) for second opinions and external reviews.
Before using Codex, verify it's installed:
which codex >/dev/null 2>&1 || echo "CODEX_NOT_INSTALLED"
If not installed, inform the user: "Codex CLI is not installed. Skipping Codex review."
OUTPUT_FILE=$(mktemp)
codex exec \
--skip-git-repo-check \
--dangerously-bypass-approvals-and-sandbox \
-o "$OUTPUT_FILE" \
"$PROMPT" </dev/null 2>/tmp/codex_stderr.log
cat "$OUTPUT_FILE"
rm "$OUTPUT_FILE"
| Flag | Purpose |
|---|---|
exec | Non-interactive mode |
--skip-git-repo-check | Run outside git repositories |
--dangerously-bypass-approvals-and-sandbox | No prompts, no sandbox restrictions |
-o <file> | Capture clean output to file |
</dev/null | Required. Close stdin — without it codex blocks on "Reading additional input from stdin..." and the -o file comes back empty (exit 0, zero bytes). Bites hardest in non-interactive / background Bash where there's no TTY. |
2>/tmp/codex_stderr.log | Send stderr to a file, not 2>/dev/null — if codex hangs or errors you need to see the stdin/auth message. Inspect the log when the output file is empty. |
Codex can take several minutes for complex prompts. Use 15 minute timeout:
timeout 900 codex exec --skip-git-repo-check --dangerously-bypass-approvals-and-sandbox -o "$OUTPUT_FILE" "$PROMPT" </dev/null 2>/tmp/codex_stderr.log
OUTPUT_FILE=$(mktemp)
codex exec --skip-git-repo-check --dangerously-bypass-approvals-and-sandbox -o "$OUTPUT_FILE" \
"What are the tradeoffs between Redis and Memcached for session storage?" 2>/dev/null
cat "$OUTPUT_FILE"
rm "$OUTPUT_FILE"
DIFF=$(git diff main...HEAD)
OUTPUT_FILE=$(mktemp)
codex exec --skip-git-repo-check --dangerously-bypass-approvals-and-sandbox -o "$OUTPUT_FILE" \
"Review this diff for issues:
$DIFF
List concerns with severity (blocking/important/nice-to-have)." 2>/dev/null
cat "$OUTPUT_FILE"
rm "$OUTPUT_FILE"
OUTPUT_FILE=$(mktemp)
codex exec --dangerously-bypass-approvals-and-sandbox -C /path/to/repo -o "$OUTPUT_FILE" \
"Analyze the architecture of this codebase" 2>/dev/null
cat "$OUTPUT_FILE"
rm "$OUTPUT_FILE"
Codex has access to gh CLI. Pass the PR URL and let it fetch the diff:
OUTPUT_FILE=$(mktemp)
timeout 900 codex exec --skip-git-repo-check --dangerously-bypass-approvals-and-sandbox -o "$OUTPUT_FILE" \
"Review this PR: https://github.com/owner/repo/pull/123
Use gh CLI to get the diff and review for issues." 2>/dev/null
cat "$OUTPUT_FILE"
rm "$OUTPUT_FILE"
/dev/null. Even with a prompt passed as an argument, codex exec reads stdin; in a non-interactive/background Bash call with no TTY it blocks ("Reading additional input from stdin...") and the -o output file ends up empty with exit code 0. The simpler usage examples above omit </dev/null — add it to every invocation.npx claudepluginhub yuvasee/samocodeConsults OpenAI Codex (GPT-5) via CLI for code investigation, debugging, or review. Runs read-only with full project access; activates on 'ask codex' phrases or /ask-codex.
Bridges Claude with OpenAI Codex CLI for delegating bulk work, reviewing uncommitted diffs, or asking read-only questions. Uses exec, review, and ask subcommands.
Cross-model review using OpenAI Codex to independently verify plans or code diffs, iterating up to 5 rounds. Useful for architecture decisions, non-trivial refactors, and critical config changes.