Menu
Run Codex tasks via the ai_cli Task Tool runner.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Set up a repo-local just task runner with a root Justfile, optional .just/ helper scripts, and curated help, build, fmt, lint, test, and ci recipes. Use when a repo needs just, a Justfile, .just helpers, or when the user mentions task automation, "just build", "just lint", "just fmt", or dropping in a just system.
Create or harden CLIs intended primarily for AI or system consumption and secondarily for humans. Use when designing or implementing a JSON-first CLI with an MCP-ready contract seam, where every command supports machine output, errors are typed and actionable, mutating commands are auditable via corresponding read commands, and external integrations need simulator-backed testing. Do not use for human-first shell utilities, one-off scripts, or text-only CLIs.
Design stateful simulators for AI-facing CLIs that integrate with devices, networks, services, or databases. Use when a CLI needs realistic simulator-backed testing, configurable fault injection, persistent state, and a swappable adapter boundary so the same business logic runs against live and simulated backends. Do not use for shallow mocks, stateless test doubles, or CLIs with no external integration.
Critically review an existing CLI, MCP wrapper, or CLI plan for AI-first contract quality. Use when evaluating whether a CLI is JSON-first, has an MCP-ready contract seam without JSON reshaping, auditably models state changes, uses stateful simulator-backed testing for external integrations, and returns typed actionable errors. Do not use for general style review or human-first shell UX review.
Run CI quality gates with optional auto-fix and PR creation.
Harden coding, QA, and orchestration agent prompts so they fix issues autonomously instead of letting findings pass through as minor, pre-existing, non-blocking, or technical debt. Use when editing existing agent prompts, review instructions, or orchestration workflows to remove permissive issue-handling behavior and tighten escalation criteria.
| name | codex-agent |
| description | Run Codex tasks via the ai_cli Task Tool runner. |
| version | 0.12.0 |
Use this skill to run Codex tasks via the ai_cli runner. This is intended for Claude to delegate work to Codex when appropriate, including long-running background tasks that can be monitored while other work continues.
Call the runner script (installed path in projects is .claude/scripts):
python3 .claude/scripts/sc_codex_task.py --json '{...}'
Use sc_codex_task.py only (do not call other runner script names).
Do not invent flags like --run_in_background, --description, --prompt, or --subagent_type.
Provide Task Tool input JSON with:
descriptionpromptsubagent_type (defaults to sc-codex if not provided by the caller)--model codex (default: gpt-5.2-codex)--model codex-max or --model max (maps to gpt-5.1-codex-max)--model codex-mini or --model mini (maps to gpt-5.1-codex-mini)--model gpt-5 (maps to gpt-5.2)--no-background is provided.--background to force background explicitly.--no-background to force blocking mode.output_file (JSONL transcript path) and agentId.output_file via a short Python loop (avoid tail -f and avoid timeout, which may be missing on macOS).--background) returns { "output", "agentId" }..claude/state/logs/<package-name>/.gpt-5.2 compatibility mode before failing.resume is provided but the prior transcript cannot be reopened or does not contain usable assistant output, return a concise error instead of inventing missing context.