jeremy
d1030def4e
Two pieces of work.
1. Keep milestones, not steps. Staged NN_*.py lanes were saving a full
~80 MB .blend per attempt, so Lena's lane reached 3.0 GB of which 2.2 GB
was 30 .blend files -- five snapshots to land one crotch fix, five more
for the bra. The .py recipes are the real history; the blends are cache.
- .agents/rules/working-files.md: four-tier policy (KEEP / MASTER /
SCRATCH / UNKNOWN) + how to work a lane.
- tools/prune_lane.py: classifies a lane and prunes the scratch tier.
Dry-run by default. Reads a per-lane .lanekeep manifest, flags
binaries byte-identical to a registered original (canonical name
always survives a duplicate pair), and never auto-deletes a .blend
with no step script beside it -- those cannot be rebuilt.
- .gitignore: scratch patterns can never be committed.
Dry run on characters/female/lena_nude reports 2.3 GB reclaimable.
Not applied -- that lane had a live Blender session at the time.
2. .humans/marvelous-designer.html: how we author garments in Marvelous
Designer, written for people rather than agents -- the six-step process,
what has been made, the traps that cost hours, and what is still
unsolved. Matches the .humans/ HTML convention in ariki-game.
Also committing the docs the AGENTS.md knowledge map and the new page
reference, so they are not dangling: the clothing-lane architecture page,
the marvelous-designer skill, and the two screenshots the page embeds.
characters/ is deliberately untracked and stays that way -- it holds GBs of
blends and GLBs, and .gitattributes does not LFS-track .blend.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
4.5 KiB
4.5 KiB
AGENTS.md — animation repo
Entry point for AI agents working in local.repo/animation (all tools: pi,
Claude Code, Cursor). This repo is Jeremy's animation-pipeline hub and the
Mac↔PC bridge that carries iClone/mocap exports into ariki-game. iClone 8 +
Video Mocap live on the PC; Blender 5, the retarget tools, and the game repo
live on both machines.
Read .agents/SOUL.md at session start, then use the knowledge map below —
this file stays thin; the operational detail lives in .claude/skills/ and
.agents/.
Knowledge map
| Category | Path | When |
|---|---|---|
| Identity | .agents/SOUL.md |
always-on |
| Batch workflow, canonical retarget command, naming, loop QC, gotchas, repo etiquette | .claude/skills/animation/SKILL.md |
always-on — read before any batch or clip work |
| Architecture (pipeline stages, canonical rig, tool provenance) | .agents/wiki/ARCHITECTURE.md |
consult when touching the pipeline |
| Per-system architecture detail | .agents/wiki/architecture/ |
consult-when |
| Clothing lane — how a reference image becomes a worn in-game outfit (both halves, and where they meet) | .agents/wiki/architecture/clothing-lane.md |
read before any garment/outfit work |
| Garment authoring in Marvelous Designer (bridge, drafting, draping, QC) | .claude/skills/marvelous-designer/SKILL.md |
always-on for garment authoring |
| Garment mesh → game-ready skinned outfit GLB (headless Blender) | clothing/README.md |
always-on for the game-ification half |
| In-flight work + recently shipped | .agents/wiki/master-plan.md |
check before starting new work |
| Dance/clip naming registry (source of truth for names) | .agents/wiki/dances/REGISTRY.md |
always-on when naming or shipping a clip |
| PC-lane routing stub | .agents/wiki/iclone-bridge.md |
consult-when |
| Point-in-time audits / convergence reports | .agents/wiki/devops-reports/ |
consult-when |
| Open/executing plans | .agents/plans/ |
consult-when |
| Other operator playbooks (iClone/mocap, clip authoring, pose estimation) | .claude/skills/ (index: .agents/skills/README.md) |
consult-when |
| Loop-QC sub-agent | .claude/agents/loop-qc.md |
consult-when |
| Repo-wide conventions | .agents/rules/ |
always-on |
Working files — keep milestones, not steps: which .blend/render dirs survive a staged lane, and how to prune the rest |
.agents/rules/working-files.md |
read before running any staged NN_*.py lane |
| Exchange-folder operator guide (which folder, when files move/archive) | exchange/GUIDE.md |
consult-when doing a batch |
| Human-facing explainer pages (HTML, for Jeremy/Özlem — not a source of truth) | .humans/ |
consult-when explaining the repo to a person |
Layout
AGENTS.md ← you are here
.agents/ ← agent context: SOUL, wiki (architecture/master-plan/registry/
devops-reports), plans, rules, skills-pointer
.claude/ ← Claude-Code-native skills + agents (auto-discovered)
tools/ ← Blender-headless retargeters, loop tools, prop export,
md_bridge* (drive Marvelous Designer), tailor/ (garment
authoring: recipes, measurements, QC, screenshots)
characters/ ← character body models: originals/ (read-only Tripo sources) +
per-character working folders (male/mako, female/lena).
Naming law in characters/REGISTRY.md — applies to all new
derived models
clothing/ ← garment mesh → game-ready outfit GLB (staged Blender pipeline)
garments/ ← raw incoming garment sources + design advice notes
exchange/ ← files in flight between Mac and PC (see exchange/GUIDE.md)
archive/ ← retired takes — never deleted, never renumbered
Two lanes run through this repo: the animation lane (motion → clips) and the clothing lane (garments → outfits). They converge on the same 65-bone Quaternius skeleton.
Git
- Branch is
main. Sync viatinqs push/tinqs pull(not raw git — this repo uses Git LFS for.fbx/.glb/.mp4/.task; a plaingit pullhangs). Report the commit hash after pushing. - Stage explicit paths, never
git add -A— ariki-game and this repo often carry unrelated in-flight files. - Never commit secrets.
The rule that bites most often
Green verify ≠ correct. An agent-authored gate once passed an upside-down
boat export. Always render and eyeball a new visual asset before shipping —
full gotcha list in .claude/skills/animation/SKILL.md.