From codex-exec
Run the Codex CLI for non-interactive coding work. Use when users ask to run `codex exec`, `codex review`, or `codex exec resume`, continue a prior Codex session, or delegate software engineering work to OpenAI Codex from the terminal.
How this skill is triggered — by the user, by Claude, or both
Slash command
/codex-exec:codex-execThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Use this skill when you need the local `codex` CLI from a terminal harness.
Use this skill when you need the local codex CLI from a terminal harness.
Bias toward non-interactive runs. Use the interactive Codex UI only when the
user explicitly wants to stay inside it.
Confirm the CLI is available before you build a workflow around it:
codex --version
If that fails, stop and report the setup issue instead of retrying the task.
Unless the user asks for something else:
codex exec for one-shot work.codex review for code-review requests.codex exec resume --last to continue the most recent saved session.gpt-5.4 as the default model.medium reasoning for ordinary work, high for harder tasks, and low for tiny checks.--sandbox read-only for codex exec analysis runs, and use codex review for review tasks.workspace-write or --full-auto only when Codex should edit files.--json or --output-schema only when another tool will parse the result.Do not hide stderr by default, and do not add --skip-git-repo-check unless you
are intentionally running outside a Git repo.
For the current flag surface and example command matrix, see
./references/codex-cli.md.
Default choice for most delegated work:
codex exec \
--model gpt-5.4 \
--sandbox read-only \
-c model_reasoning_effort="medium" \
"Summarize the uncommitted changes"
Use the dedicated review subcommand instead of hand-rolling a review prompt
through exec:
codex review --uncommitted \
"Focus on bugs, regressions, and missing tests. Findings first."
Use stdin when the follow-up prompt is large or multi-line:
printf '%s\n' "Continue and focus on the failing tests only" | \
codex exec resume --last -
When resuming, keep prior settings unless the user explicitly wants to change model, sandbox, or autonomy level.
codex review is the default review path when the task is code review.codex exec reads instructions from stdin when the prompt argument is omitted or set to -.codex exec resume --last is the non-interactive continuation path; do not replace it with the top-level interactive codex resume command.--full-auto is only a convenience alias for editable autonomous runs. It is not appropriate for read-only analysis.--skip-git-repo-check is a situational escape hatch, not a default.--ephemeral is useful for disposable runs when session persistence would add noise.Use structured output only when another tool needs to consume the result:
codex exec \
--model gpt-5.4 \
--sandbox read-only \
--output-schema schema.json \
"Review the current diff"
codex --version fails, report a CLI issue.codex exec or codex review exits non-zero, treat the run as failed.codex review before inventing a custom exec prompt.On first use in a session, silently check for a newer version:
https://raw.githubusercontent.com/abpai/skills/main/versions.json.codex-exec against this file's metadata.version.codex-exec update available (local {X.Y} → remote {A.B}). Would you like me to update it for you first? I can run
npx skills update codex-execfor you.
Provides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub abpai/skills --plugin codex-exec