메뉴
Use when the user wants to design Codex agent equivalents (specialized workers/profiles/prompt files), define triggering conditions, or build reusable agent prompts and validation tools.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Use before starting ANY work - hard gate ensuring a GitHub issue exists, creating one if needed through user questioning
Use when receiving UAT feedback, bug reports, user testing results, stakeholder feedback, QA findings, or any batch of issues to investigate. Investigates each item BEFORE creating issues, classifies by type and priority, creates well-formed GitHub issues with proper project board integration.
MANDATORY for all work - the project board is THE source of truth. This skill provides verification functions and gates that other skills MUST call. No work proceeds without project board compliance.
Use for LARGE work requiring feature-level grouping. Creates epic tracking issues, manages related issues under a common label, tracks epic progress, and coordinates with milestones.
Create distinctive, production-grade frontend interfaces with high design quality. Use this skill when the user asks to build web components, pages, or applications. Generates creative, polished code that avoids generic AI aesthetics.
Use when the user wants to create Codex workflow hooks (pre/post run gates, tool-use validators, stop checks) or needs guidance on hook scripts and hooks.json configuration.
| name | agent-development |
| description | Use when the user wants to design Codex agent equivalents (specialized workers/profiles/prompt files), define triggering conditions, or build reusable agent prompts and validation tools. |
Codex CLI does not auto-discover "agent" files the way Claude plugins do. To recreate agent behavior in Codex, combine:
codex-subagent, codex exec, or worker-dispatch) for isolated execution.This skill standardizes an agent card format (YAML frontmatter + prompt body) and shows how to run it with Codex.
Store agent prompts in agents/<agent-name>.md. Frontmatter is metadata for humans and validators; Codex does not read it automatically.
Required fields:
name (kebab-case identifier)description (when to use this agent)Recommended fields:
profile (Codex config profile name)model (override model)sandbox_mode and approval_policy (safety defaults)tools (advisory list of tools; use sandbox/approval to enforce)Example:
---
name: pr-reviewer
description: Use this agent when the user asks for a code review or to audit a pull request.
profile: review-worker
model: gpt-5.2-codex
sandbox_mode: workspace-write
approval_policy: on-request
tools: ["Read", "Grep", "Bash"]
---
You are a senior PR reviewer...
Preferred (subagent runner):
codex-subagent pr-reviewer <<'EOF'
[Provide task context and scope here.]
EOF
Fallback (manual codex exec):
awk 'BEGIN{c=0} /^---$/{c++; next} c>=2 {print}' agents/pr-reviewer.md | codex exec -p review-worker -
Define per-agent profiles in ~/.codex/config.toml:
[profiles.review-worker]
model = "gpt-5.2-codex"
sandbox_mode = "workspace-write"
approval_policy = "on-request"
Profiles capture the model and safety settings that Claude agents used to embed in frontmatter.
To make agent usage automatic, create a skill describing the trigger and invocation. The skill body should instruct how to run the agent card or spawn a worker.
Example trigger entry:
description: Use when the user asks for a PR review or code audit; run the pr-reviewer agent card.
See references/triggering-examples.md.
Use scripts/validate-agent.sh to validate agent cards for required fields and prompt length.
references/system-prompt-design.md for prompt structurereferences/triggering-examples.md for trigger phrasingreferences/agent-creation-system-prompt.md for AI-assisted agent card generationexamples/agent-creation-prompt.md for a complete prompt templateexamples/complete-agent-examples.md for full agent card examples