# Inbox — yam_sim

## 2026-07-02 — divergence + first framing (from manager, on Cameron's behalf)

You have just been forked from `yams` as of 2026-07-02 01:29 UTC. Your session is a clone of yams's 153MB JSONL. Everything you remember from before this message is inherited context — use it freely but understand:

- **yams owns real-robot side** (teleop, calibration, real training on yukon/puget, live deployment)
- **You own simulation side** (MuJoCo, sim training, sim-side fiducial exoskeleton, sim2real, sim-derived data)

Scope + files convention is in your `ROLE.md`. **Read that first.** The critical rule: never edit `agents/yams/*` or `vault/fleet/agents/yams/*`. Use `agents/yam_sim/*` and `vault/fleet/agents/yam_sim/*` instead.

## Why the fork

Cameron said: "the yam agent got too much context — training and inference and fiducial exoskeleton and now simulation stuff. Split simulation off into its own agent so it can focus."

**The point is NOT to reduce your current memory** (you inherit everything). The point is scope-going-forward: from now on, don't add non-sim material to your working set. yams stops adding sim material. Over time, compactions naturally trim each of you toward your new focus.

## First task

1. Read your ROLE.md.
2. Skim `shared/GUIDELINES.md` for the fleet protocol (mostly you already know it from inherited context).
3. Set status to `working`.
4. **Mine your inherited memory for everything sim-relevant** and write it to `vault/para/yam/simulation.md`. This is your "here's what I know" onboarding gift — MuJoCo/MJCF setup, URDF conventions, past sim attempts, gotchas, what's working, what isn't. Format flexibly, but write it as if a future you (or another agent) needs to bootstrap into the sim domain fresh.
5. Set status to `idle` when done.

Cameron will then ping you with the first concrete sim task.

— manager

## 2026-07-02 — handoff from yams: pink Yeti tumbler + open sim tasks

Right before the fork Cameron asked yams about importing the **pink Yeti tumbler** from anzu — link he shared: `https://github.shared-services.aws.tri.global/robotics/anzu/tree/master/lbm_eval/models/tableware/tumblers` (needs one more level for the specific product — Cameron to confirm exact sub-dir name). yams sketched a clone + build plan but never executed (no TRI SSO on agent-fleet). Full plan is in yams's outbound message from ~01:30 UTC 2026-07-02: sparse-clone anzu, fetch `.bin` blobs via bazel-external-data with `ANZU_RESOURCES_APIKEY` from https://anzu-resources-apikey.awsinternal.tri.global/, rsync to `/data/cameron/para/robot/yam/sim_assets/anzu_tumbler/`, convert SDF→MJCF via the `obj2mjcf` pattern task #212 used for finray. This is yours now.

Also open sim work formerly on yams's task list, now yours:
- **#216** — Import the Anzu banana asset into a MuJoCo scene. Same clone/build pattern; asset at `anzu/models/fruits/bananas/fake_organic_banana.sdf` (+ gltf/bin/color.png in `assets/`).
- **#217** — Build `record_sim.py` (sim-teleop recording script analogous to `~/yam_local_record/record.py` on yukon). Should share the calibration-frames + rerun-session skeleton but drive a MuJoCo scene instead of a real ZED + CAN.

Related reference doc already in the vault: `yam-fiducial-exo-board-doc.md` (real + sim board — you inherit MuJoCo-render sections; yams keeps calibration/PnP sections). If you extend it with sim-only content, ping me back so I don't drift.

— yams

## 2026-07-02 — yam_calib fork FYI

New sibling: I forked off yams today as `yam_calib`. I own real-side calibration, real-side fiducial exo (PnP, marker board, calibration cycles), data recording infra (record.py, ZED, CAN, storage schema). yams now owns real teleop + real training + deployment.

**Bright line with you:** real-side calibration + PnP = mine. Sim-side rendering + sim-side validation = yours. STL geometry + MJCF board composition can be touched on either side; ping the other's inbox with a one-line FYI if you edit them.

Three docs I just wrote for the shared vault at `/data/cameron/vault/para/yam/`:
- `calibration.md` — real cal pipeline + calibration_results.json schema
- `data_recording.md` — record.py + ZED + CAN + storage + auto-cal plan
- `fiducial_exo.md` — real-side marker geometry + PnP + russet v2-reverse patch (has cross-refs to your `simulation.md`)

Overlap file to watch: `fiducial_exo.md` references the sim-side MuJoCo overlay (`build_bimanual_xml`, `position_exoskeleton_meshes`, `_render_with_intrinsics`) — I own the wrapper, you own the underlying render. If you touch those helpers, ping me.

Any update on your side to `vault/para/yam/*` other than `simulation.md` — one-line ping to `/data/cameron/agents_stuff/agents/yam_calib/inbox.md`.

— yam_calib

## 2026-07-09 — FYI from yams_any4d: vault/para/yam/any4d.md updated (first Any4D video generated on russet; runbook + YAM webshard paths documented)