assetforge
/assetforge
Generuje kandydatów na zasoby gier w ramach limitu kosztów i przygotowuje je do wyboru przez człowieka.
- Autor
- clockgrove ↗
- Repozytorium
- assetforge ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 13.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 1
Brak sygnałów. Repozytorium zmienione 13.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Generuje kandydatów na zasoby gier w ramach limitu kosztów i przygotowuje je do wyboru przez człowieka.
Oryginalny opis (EN)
Use when you need to generate candidate game assets within a cost cap and stage them, with QA and provenance, for a human to pick.
Jak działa
- Contract
- Generation plans
- Commands
- Exit codes
- Environment and network
- The rule (PRD section 5.7)
- Suggested flow
Podgląd SKILL.md
---
name: assetforge
description: Use when you need to generate candidate game assets within a cost cap and stage them, with QA and provenance, for a human to pick.
---
# Assetforge
Assetforge generates game assets through generation models and stages the ones that pass QA, with provenance, for a human to review.
Run it as `npx @clockgrove/assetforge@<version> <command>`, or as `assetforge` (`pnpm exec assetforge`) when the game pins it as a dev dependency.
## Contract
- Always pass `--json`. Stdout is exactly one JSON document with `ok`, `exitCode`, `command` and, on refusal, `error: { code, message }`.
- Logs go to stderr. It never prompts.
## Generation plans
For repository-owned art direction, initialize the target's constitution with
`bible init --directory <target> --world <name> --preset <preset> --json`.
Load `bible describe --bible <BIBLE.md> --json` before writing a structured brief.
The presets are `cozy-storybook`, `flat-cartoon-kids` and `dark-gritty-isometric`.
Switch look and tone with `bible preset --bible <BIBLE.md> --preset <preset> --json`;
it retains the world palette, anchors and identity locks and invalidates earlier plans.
Use `plan --spec <spec.json> --bible <BIBLE.md> --profile <profile.json>
--brief <brief.json> --out <plan.json> --json` to compile a reviewable plan without
generating images. Show its prompts, clamps and cost forecast to the human.
`apply <plan.json> --json` requires the human to set the plan's `approved` field;
an unattended campaign uses `--auto-approve` to authorize generation spend.
This approval never approves an asset. Apply refuses stale bible, template,
identity and reference inputs. `mcp` exposes the same flow over stdio.
See [generation knobs](https://github.com/clockgrove/assetforge/blob/main/docs/generation-knobs.md) in the source repository
for the brief, run settings and versioned target template formats.
## Commands
- `stage --spec <spec.json> --bible <bible.json> --profile <profile.json> [--candidates 4] [--cap 5] [--routes <routes.json>] [--runs .assetforge/runs] [--provider-script <module.js>] [--dry-run] [--json]`
Generates candidates within the USD cost cap and stages only those passing QA, with reasons and provenance, in a run directory. `--dry-run` reports the planned calls and cost forecast and spends nothing (needs no key).
- `pick --run <dir> --candidate <id> --target <dir> [--json]`
Writes the chosen candidate and manifest into the target with review pending.
- `rig --model <model.glb> --provenance <provenance.json> --spec <spec.json> --profile <profile.json> [--runs .assetforge/runs] [--json]`
Re-rigs an existing 3D character on the target's shared rig with generated gaits, checks every clip, and stages it as a one-candidate run for a human to pick. Calls no model and spends nothing.
- `look-dev --model <model.glb> --provenance <provenance.json> --spec <spec.json> --bible <bible.json> --profile <profile.json> [--cap 3] [--routes <routes.json>] [--runs .assetforge/runs] [--work-item <label>] [--dry-run] [--json]`
Repaints a picked 3D character's base colour in the bible's character look on its own UVs, writes fur groom maps (and tints) per season the profile declares, gives the eyes a glossy material and mask, checks them, and stages a one-candidate run with before-and-after renders for a human to pick. Never approves.
- `import --sheet <sheet.png> --profile <profile.json> --asset <id> --tool <name> [--bible <bible.json>] [--hints idle,walk-right] [--frames <count>] [--grid <columns>] [--components] [--no-propose] [--cap 5] [--provider-script <module.js>] [--json]`
Slices a contact sheet into frames and has a vision model propose animation families in `<import>/grouping.json` with status `proposed` (`--no-propose` calls no model). With `--frames`, a sheet that does not slice into exactly that many figures is refused (exit 2), naming the cells behind it.
- `generate-sheet --spec <spec.json> --bible <bible.json> --profile <profile.json> [--brief <brief.json>] [--cap 5] [--provider-script <module.js>] [--dry-run] [--json]`
Has the model draw a whole cycle as one sheet within the cap, then imports it. A sheet that does not slice into the spec's `frames.count` figures exits 5; the sheet and its record are kept in the import directory.
- `write-families --import <dir> --target <dir> [--grouping <file>] [--json]`
Writes the strips of a grouping a human set to `confirmed` into the target with review pending.
- `op <operation> <input...> --out <path> [--params <json>] [--profile <profile.json>] [--json]`
Runs one image operation (`assetforge ops` lists them with their parameter schemas) on any image and writes the output and a provenance record. Local, with no model call and no spend.
- `recipe <recipe.json> <input...> --out <path> [--profile <profile.json>] [--json]`
Runs a recipe of operations deterministically. Same input, recipe and engine versions, same bytes.
- `ops [--json]`
Lists the operations, their parameter schemas and the engines.
- `--help`, `--version`; `<command> --help` for one command's options. With `--json`, help is a JSON document whose `usage` field holds the text.
## Exit codes
| Code | Meaning |
| ---- | ---------------------------------------------------------------------------------------------------------------------------- |
| 0 | Success |
| 1 | Unexpected internal error |
| 2 | Invalid input |
| 3 | Cap refused: the run's cap or a spend limit across runs (also a dry-run forecast over either) |
| 4 | Provider error (missing `AI_GATEWAY_API_KEY`, or every candidate failed) |
| 5 | QA found nothing |
| 6 | The target's validator rejected what was written |
| 7 | Licence refused (`license_refused`): the provider's output grant does not let the target declare the output; nothing written |
## Environment and network
- Reads `AI_GATEWAY_API_KEY` (Vercel AI Gateway key), from the environment only. Not needed for `--dry-run` or `--provider-script`.
- Records every paid call in a spend ledger shared across runs (`ASSETFORGE_SPEND_LEDGER`, default `~/.local/state/assetforge/spend.jsonl`) and refuses a call (exit 3) that would cross the daily limit (`ASSETFORGE_DAILY_LIMIT_USD`, default $10) or the work item's limit (`ASSETFORGE_WORK_ITEM_LIMIT_USD`, default $5). Name your work item with `ASSETFORGE_WORK_ITEM` or `--work-item`, for example the Objective; in a Factory worker it defaults to the Objective from `FACTORY_ISSUE`. `--dry-run` shows the spend left under each. Scripted runs (`--provider-script`) are not counted or refused by these limits; their `--cap` still applies. Never raise a limit yourself; ask the human.
- Reaches only `ai-gateway.vercel.sh` (HTTPS), and only for `plan`, `apply`, `stage`, `generate-sheet` and `import` (without `--no-propose`), or their MCP equivalents, never with `--dry-run` or `--provider-script`. `pick` and `write-families` run the target profile's own validator command locally. A sandbox can allow just that host.
## The rule (PRD section 5.7)
**Agents may plan, apply authorized generation plans, stage candidates, generate and import sheets (and dry-run first to check cost). Agents never run `pick`, never confirm a grouping or run `write-families`, and never approve. A human picks, and Assetforge never marks anything approved.** Report the staged run or import directory and the candidate IDs to the human instead.
## Suggested flow
1. `stage --dry-run --json` to check the plan and cost forecast.
2. `stage --json` to generate and stage candidates.
3. Hand the run directory and candidate IDs to the human.
Wersje
W bazie nie ma innych skilli o nazwie assetforge.
Migawka bazy z 13.11.2026 · wersja paczki a7317ebac762 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100assetcat
Wyszukuje zasoby do gier, sprawdza licencje i kopiuje je do projektu z indeksowanej biblioteki przez CLI assetcat. - 95game-development
Wspiera tworzenie gier w Unity, Unreal Engine i Godot: mechaniki, fizykę i AI w silnikach gier. - 84review
Prowadzi etap przeglądu w Fabriku: sprawdza implementację, naprawia problemy i przygotowuje PR do przeglądu przez człowieka. - 81game-studio
Konfiguruje Codex Game Studio dla nowej lub istniejącej gry i kieruje do właściwych przepływów pracy twórcy gier. - 100image-generation
Generuje zasoby graficzne do gier przez API Google Gemini i przetwarza je w gotowe pliki sprite'ów, z wyborem modelu, proporcji, obrazu referencyjnego i liczby wariantów.