copet-gen

name	copet-gen
description	Use when generating a custom pet for the CoPet app from a concept, image, brand, mascot, or visual prompt, and the finished hatch-pet package must be installed into $HOME/.copet/pets with collision-safe pet ids.

CoPet Gen

Overview

Create one CoPet pet package under $HOME/.copet/pets/<pet-id>/ by delegating pet generation to $hatch-pet, then copying the finished package into CoPet's app config directory.

This Skill does not generate sprites, assemble atlases, run visual QA, or repair pet rows directly. $hatch-pet owns the full pet generation workflow.

Using $copet-gen is explicit permission to use the subagents and lightweight workers required by $hatch-pet. Do not downgrade to main-session sequential generation solely because $hatch-pet delegates visual jobs.

Codex only. copet-gen delegates pet generation to $hatch-pet, which in turn delegates image generation to $imagegen. $imagegen's default path is Codex's built-in image_gen tool, so this Skill must run inside Codex. Claude Code, Cursor, and other agents do not ship the required tool and are not supported.

Required Dependency

$copet-gen requires $hatch-pet (source: https://github.com/openai/skills/tree/main/skills/.curated/hatch-pet). The workflow installs $hatch-pet automatically when it is missing from the active agent's skill directory — see Workflow step 1. Do not substitute direct $imagegen calls, local image scripts, hand-authored sprites, or partial pet package generation.

If the local agent framework requires explicit user permission before spawning subagents, treat a user request to use $copet-gen as that permission for $hatch-pet pet-generation workers, visual QA workers, and brand-discovery workers.

Package Layout

$HOME/.copet/
└── pets/
    └── <pet-id>/
        ├── pet.json
        └── spritesheet.webp

If <pet-id> already exists under $HOME/.copet/pets/, append -2, -3, and continue until the destination is unique. The copied package's pet.json.id must exactly match the installed directory name.

Workflow

1. Ensure $hatch-pet is available in the active agent. If it is not registered, install it:

npx skills add openai/skills --skill hatch-pet

After installation, reload the agent so $hatch-pet becomes invocable, then continue.

2. Use $hatch-pet to complete the user's pet request.

   • Wait for $hatch-pet to finish validation, visual QA, repair if needed, and packaging.
   • The expected source package is usually ${CODEX_HOME:-$HOME/.codex}/pets/<pet-id>/ (this path is $hatch-pet's output directory and is independent of which agent invoked it).
   • If the source package path is unclear, read the hatch run's qa/run-summary.json and use its package field.

3. Install the completed package into CoPet:

python skills/copet-gen/scripts/install-copet-pet.py \
  --source-pet-dir "${CODEX_HOME:-$HOME/.codex}/pets/<pet-id>"

By default the installer writes to ${COPET_CONFIG_DIR:-$HOME/.copet}/pets. If the user supplies a different CoPet config directory, pass it explicitly:

python skills/copet-gen/scripts/install-copet-pet.py \
  --source-pet-dir /absolute/path/to/source-pet \
  --copet-config-dir /absolute/path/to/.copet

4. Verify the installed package:

INSTALLED_DIR=/absolute/path/from/installer-output
test -f "$INSTALLED_DIR/pet.json"
test -f "$INSTALLED_DIR/spritesheet.webp"
test "$(basename "$INSTALLED_DIR")" = "$(jq -r '.id' "$INSTALLED_DIR/pet.json")"

5. Report installed_pet_id and installed_pet_dir. If collision=true, mention the assigned suffix.

Rules

• Auto-install $hatch-pet when missing instead of stopping — only halt if npx skills add itself fails.
• Always complete $hatch-pet before installing into CoPet.
• Allow $hatch-pet to spawn its normal subagents and lightweight workers.
• Copy from the $hatch-pet package; do not move or mutate the source package.
• Use scripts/install-copet-pet.py for installation instead of ad hoc copy commands.
• Keep the installed directory name and copied pet.json.id identical.