# Create Custom Prompt

> Create Codex custom prompts (slash commands) by generating Markdown prompt files under `~/.codex/prompts/` with YAML front matter (`description`, optional `argument-hint`) and placeholders, then open the prompt in VS Code. Use when a user asks to create/update a reusable `/prompts:<name>` command for Codex CLI or the Codex IDE extension.

- Skill: `itd-cba/create-custom-prompt` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add itd-cba/create-custom-prompt`
- Raw SKILL.md: https://api.skillmd.com/api/skills/itd-cba/create-custom-prompt/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: itd-cba (https://skillmd.com/u/itd-cba)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/itd-cba/create-custom-prompt

---


# Create Custom Prompt

## Workflow

Create a new Codex custom prompt file in `~/.codex/prompts/` following the Codex custom prompts format, then open it in VS Code.

### 1) Gather requirements (ask first)

Ask the user what they want to achieve with the prompt, then ask only the minimum clarifying questions needed to generate a usable prompt:

- **Goal**: What should the slash command accomplish, and what “done” looks like?
- **Scope**: Is the prompt repo-specific (e.g. this microfrontend) or general-purpose?
- **Inputs / arguments**:
  - Do they want **named** args (`FILES=...`, `TICKET_ID=...`) or **positional** args (`$1`, `$2`, …)?
  - Which args are required vs optional?
  - Any quoting expectations (paths with spaces, multi-line text, etc.)?
- **Output style**: concise vs detailed; include steps/checklists; include code blocks?
- **Safety constraints**: anything the prompt must never do (e.g., “don’t run destructive commands”)?

If the user is unsure about arguments, propose a small set of named placeholders (uppercase) and confirm.

**Make it more concrete (recommended):** If you have ideas that materially improve the prompt’s clarity or reliability, surface them as suggestions and confirm with the user. Keep this tight (1–5 bullets) and focus on high-leverage concreteness, for example:
- Add/adjust arguments to avoid ambiguous input (e.g., `FEATURE_NAME=...`, `PLAN_FILE=...`, `OUT=...`, `SCOPE_DIR=...`).
- Nail down output artifacts and filenames (where to write results, overwrite vs append).
- Specify required output structure (sections/headings) and verification criteria.
- Add explicit “blocking questions only” vs “ask first” behavior when appropriate.
- Add tool/command expectations (e.g., use `git diff`, run `pnpm lint`) and guardrails (“no new tooling unless requested”).

### 2) Choose a command/file name

Pick a short, memorable, **lowercase** name (letters/digits, optionally hyphens). It becomes:

- File: `~/.codex/prompts/<name>.md`
- Slash command: `/prompts:<name>`

If the user did not specify a name, propose one derived from the goal (e.g. `draftpr`, `summarize-log`, `review-diff`) and confirm.

### 3) Create the prompt file

Follow the Codex custom prompts guidance (see `references/custom-prompts.md`):

- Ensure the prompts directory exists: `~/.codex/prompts/` (create it if missing).
- Create the prompt file directly under that directory (no subfolders).
- Add YAML front matter:
  - `description`: short, user-facing summary
  - `argument-hint` (optional): a compact “usage” string that matches the placeholders
- Write the prompt body in Markdown and include placeholders as needed (e.g. `$FILES`, `$TICKET_ID`, `$1`, `$ARGUMENTS`).

Prefer using the helper script so directory creation, naming, and opening are consistent:

```bash
cat <<'EOF' | scripts/create_prompt.sh \
  --name "<name>" \
  --description "<one-line description>" \
  --argument-hint "[OPTIONAL_ARG=<value>] [OTHER_ARG=\"<value>\"]" \
  --open
<prompt body markdown here>
EOF
```

This path is intentionally relative to the skill folder so it works whether the skill is installed globally (under a Codex home) or locally (inside a repository).

If the user wants to update an existing prompt, edit the existing `~/.codex/prompts/<name>.md` instead of creating a new one.

### 4) Open in VS Code

Always open the created/updated prompt file via the VS Code CLI if available:

- Use `code -g "<path>:1"`
- If `code` is unavailable, print the full path and tell the user to open it manually.

### 5) Remind about reload

Remind the user: after adding/editing custom prompt files, restart Codex (new CLI session; reload IDE extension) so it loads the updated prompt metadata and placeholders.