Add skill: codex-pet-generator

skills/.curated/codex-pet-generator/LICENSE.txt

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Daniil Makeev (wyddy7)
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

skills/.curated/codex-pet-generator/SKILL.md

@@ -0,0 +1,177 @@
+---
+name: "codex-pet-generator"
+description: "Create Codex desktop pets from character references by defining row semantics, generating tiny-UI-readable loops, approving or rejecting rows, following slicing guidance, packing a 1536x1872 spritesheet with the bundled packer, validating it, and installing under ~/.codex/pets."
+license: "MIT"
+---
+
+# Codex Pet Generator
+
+Use this skill when the user wants to create, revise, validate, package, or install a custom Codex desktop pet.
+
+## Core Principles
+
+- Animation direction first, image generation second, deterministic packing third.
+- Optimize for tiny UI readability, not for how pretty a large strip looks in isolation.
+- Treat each row as one loopable state machine, not eight poster poses.
+- Keep canon anchors: one approved scale reference, one seated-body reference, one standing-body reference, and one prop-language reference.
+- Every generated row must be marked `approved` or `unapproved` before it can advance.
+
+## Codex Format
+
+Custom pets live at:
+
+```text
+~/.codex/pets/<pet-id>/
+  pet.json
+  spritesheet.png|webp
+```
+
+Required spritesheet:
+
+- `1536x1872`
+- `8x9` grid
+- `192x208` per frame
+- transparent final background
+- PNG or WebP
+
+Required `pet.json`:
+
+```json
+{
+  "id": "<pet-id>",
+  "displayName": "<Display Name>",
+  "description": "<one-line description>",
+  "spritesheetPath": "spritesheet.webp"
+}
+```
+
+`spritesheetPath` is resolved relative to the `pet.json` file. Use the same `<pet-id>` as the parent directory name.
+
+Row order is fixed:
+
+1. `idle`
+2. `running-right`
+3. `running-left`
+4. `waving`
+5. `jumping`
+6. `failed`
+7. `waiting`
+8. `running`
+9. `review`
+
+For row-by-row semantics, common failure modes, and tiny-UI design guidance, read [references/row-semantics.md](references/row-semantics.md) before prompting.
+
+## Workflow
+
+1. Gather references and write a character bible in plain language.
+2. Define all 9 row semantics before generating anything.
+3. Pick canon anchors:
+   - one approved overall scale reference
+   - one approved seated reference
+   - one approved standing reference
+   - one approved prop-language reference
+4. Generate one row strip at a time, or individual frames. Do not generate a full 72-frame sheet as the final asset.
+5. Review each row with the acceptance ladder in [references/acceptance-checklist.md](references/acceptance-checklist.md).
+6. Mark every candidate `approved` or `unapproved`. Only approved rows advance.
+7. Extract individual frames into per-row directories. The skill provides written slicing guidance in [references/packing-and-slicing.md](references/packing-and-slicing.md), but the actual extraction is manual — either export frames directly from the generator, or cut strips with your image tool of choice. The bundled packer assumes per-frame inputs and does not slice strips for you.
+8. Pack with `scripts/pack_codex_pet.py`.
+9. Validate with `scripts/validate_codex_pet.py`.
+10. Install into `~/.codex/pets/<pet-id>`.
+11. When iterating in the Codex UI, use a new pet id such as `<pet-id>-v2` to avoid cached spritesheets.
+
+## Prompting Rules
+
+Prompts should explicitly state:
+
+- one row equals one loopable action
+- frame `8` must hand off cleanly to frame `1`
+- frames are sequential animation states, not unrelated poster poses
+- same character, same camera, same scale across the row
+- exactly two arms and exactly two legs
+- no cropped body parts
+- no props outside the row-specific action
+- final target is a tiny desktop pet sprite, so readability beats nuance
+
+If a row is hard to read:
+
+- do not merely “polish” it
+- redesign the motion around bolder state changes
+- prefer binary readable mechanics such as `open/closed`, `upright/flat`, `connect/fail`, `tension/release`, or `in/out`
+
+High-risk guidance:
+
+- Standing rows are high-risk for vertical stretch. Always anchor them to an approved compact standing canon.
+- Directional rows often collapse into generic running. Reject them if the intended action semantics vanish.
+- Gestures often fail when they rely on finger nuance alone. If needed, redesign them into chunkier hand states.
+- Failure rows often become generic sadness. Prefer unmistakable failure silhouettes or prop logic.
+- Jump rows often become athletic. Heavy toy-like characters usually need compression, shock, or flop logic instead.
+
+## Background and Alpha Rules
+
+- Prefer generating with a removable flat background when possible.
+- If you need chroma key, use flat `#00ff00`, no shadows, no gradients, and no green on the character.
+- Remove the background locally before packing.
+- Zero the RGB channels of fully transparent pixels before final install artifacts if your cleanup leaves preview garbage.
+
+## Approval Gate
+
+Do not carry ambiguous rows forward.
+
+- `approved`: chosen final candidate for that row
+- `unapproved`: rejected, superseded, or still exploratory
+
+Only approved rows should be sliced, packed, validated, and installed.
+
+The acceptance ladder and rejection patterns live in [references/acceptance-checklist.md](references/acceptance-checklist.md).
+
+## Slicing and Packing Rules
+
+Slicing is **manual** — the skill provides guidance, not an automated slicer. The packer takes per-frame inputs and assembles them into the final sheet using one uniform scale per row.
+
+- Do not assume a generated row strip is cut into 8 equal visual columns.
+- Do not slice approved strips into equal widths unless the spacing is known to be uniform.
+- Prefer per-frame export or gap-aware manual slicing based on real empty space between poses.
+- The packer uses one scale per row, not one scale per frame.
+- Keep at least `8-12px` of padding from cell edges (packer default is `10`, hard cap is half-cell-minus-one).
+- Keep exactly one directory per row index inside the rows root. Multiple matching directories (`01-idle-explore`, `01-idle-final`) cause the packer to abort instead of silently picking one — this protects the approval gate.
+
+See [references/packing-and-slicing.md](references/packing-and-slicing.md) for manual slicing workflow guidance.
+
+## Bundled Scripts
+
+Pack approved rows into a final sheet:
+
+```bash
+python scripts/pack_codex_pet.py <rows-root> <output.webp>
+```
+
+Validate the final sheet:
+
+```bash
+python scripts/validate_codex_pet.py <spritesheet.webp>
+```
+
+If `pet.json` sits next to the spritesheet, the validator picks it up automatically and checks the install contract (`id`, `displayName`, `description`, `spritesheetPath`). To validate against a `pet.json` in a different location use `--pet-json <path>`. To skip pet.json entirely use `--no-pet-json`.
+
+Both scripts require `Pillow`. Install with `pip install -r requirements.txt`. Run from the bundle root, or use absolute paths.
+
+Interpretation rule:
+
+- validator `errors` are hard blockers — wrong dimensions, empty frames, missing alpha, fully-opaque sheet, malformed `pet.json`
+- validator `warnings` still require visual review — fully-opaque single frame, edge contact, width/height drift, `spritesheetPath` mismatch
+- width or height drift warnings may reflect real performance changes, but edge-contact warnings usually mean crop risk
+
+## Install Notes
+
+- Install final assets under `~/.codex/pets/<pet-id>/`
+- Keep `pet.json` and the spritesheet in the same folder
+- Prefer a new id when testing revisions in the UI
+
+If the UI looks broken, first suspect:
+
+- bad row semantics
+- bad slicing
+- bad alpha cleanup
+- bad packing
+
+Do not assume the generator is the only failure point.

skills/.curated/codex-pet-generator/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "Codex Pet Generator"
+  short_description: "Build Codex pets with safe row workflows"
+  default_prompt: "Use $codex-pet-generator to turn these character refs into a validated Codex desktop pet."

skills/.curated/codex-pet-generator/references/acceptance-checklist.md

@@ -0,0 +1,124 @@
+# Acceptance Checklist
+
+Use this file when reviewing generated candidates and deciding what advances to packing.
+
+## Rule Zero
+
+Do not debate polish until the row passes semantics.
+
+If the row does not read correctly at tiny UI size, reject it before discussing detail cleanup.
+
+## Acceptance Ladder
+
+Review rows in this order:
+
+1. Semantic read
+2. Loop closure
+3. Scale match
+4. Anatomy
+5. Prop clarity
+6. Packing safety
+
+## 1. Semantic Read
+
+Ask:
+
+- does this row clearly express the intended action?
+- does it still read when mentally shrunk down?
+- is the motion built around strong state changes instead of vague nuance?
+
+Reject if:
+
+- the row tells the wrong story
+- the row feels like eight poster poses
+- the row only works when viewed large
+
+## 2. Loop Closure
+
+Ask:
+
+- does frame `8` hand off naturally to frame `1`?
+- is this a cycle rather than a one-way process shot?
+
+Reject if:
+
+- the loop visibly snaps
+- the motion is narrative instead of cyclical
+
+## 3. Scale Match
+
+Ask:
+
+- does this still feel like the same physical figurine?
+- does it match the approved scale canon for seated or standing rows?
+
+Standing-row rule:
+
+- standing rows are high-risk by default
+- reject rows that grow taller, longer-legged, or more human-like than the approved compact canon
+
+Reject if:
+
+- perceived leg length increases
+- overall toy height increases
+- seated and standing rows no longer look like the same object
+
+## 4. Anatomy
+
+Ask:
+
+- exactly two arms?
+- exactly two legs?
+- no duplicate hands, feet, or merged limbs?
+- no cropped essential body parts?
+
+Reject if:
+
+- anatomy errors exist
+- the “fix” would rely on cropping them away
+
+## 5. Prop Clarity
+
+Ask:
+
+- does the prop help the action read?
+- is the prop readable at tiny size?
+- is the prop design simple enough?
+
+Reject if:
+
+- the prop blends into the costume
+- the prop becomes a blob
+- the row requires detail that will vanish in the app
+
+## 6. Packing Safety
+
+Ask:
+
+- will this row survive slicing?
+- do limbs, hair, or props extend into neighboring visual gaps?
+- is the spacing uniform enough for simple slicing, or does it need gap-aware extraction?
+
+Reject or rework if:
+
+- important parts sit on would-be cut lines
+- the strip depends on equal-width slicing when the spacing is visibly uneven
+
+## Approved vs Unapproved
+
+Every candidate must end up in one bucket:
+
+- `approved`: final chosen candidate for that row
+- `unapproved`: rejected, exploratory, superseded, or unresolved
+
+Only approved rows should be packed.
+
+## Common Reject Patterns
+
+- tiny-UI unreadable gesture
+- standing row stretch
+- directional row collapsing into generic running
+- jump row becoming too athletic or floaty
+- failure row becoming generic sadness
+- prop clutter overpowering the character
+- frame spacing that will cut off limbs during slicing