:Release Notes: - Agents files for opencode and claude - Skills for opencode and claude - 3d models with medical organs - Some textures :Detailed Notes: - :Testing Performed: - :QA Notes: - Looks like shit :) :Issues Addressed: TG-1
4.6 KiB
4.6 KiB
Change Reporting & Handoff (project-scoped)
Every bounded change and every cross-agent hand-off must use this format. Short, scannable, honest about what was not done.
When to use
- After any bounded write (material tweak, hierarchy edit, Volume override, asset import, script change).
- After any read-only inspection that the next agent will act on.
- After any refusal, rollback, or "stop, ask" decision.
- At the end of any agent's turn.
The handoff format
Use exactly these section headers, in this order. One or two short lines per section.
ROLE: <which agent produced this handoff>
OBSERVED: <what the inspection found, with concrete paths / names>
CHANGED: <exact objects / assets / settings / scripts modified, with paths>
NOT CHANGED: <relevant things intentionally left alone, including off-limits items>
RISKS: <visual, technical, performance, safety, or uncertainty notes>
NEXT SAFE OPTIONS: <one-sentence suggestions for the next agent>
Section rules
- ROLE — the agent's name (
Scene Director,Unity Execution,Materials/Assets,Lighting/Rendering,Medical-Simulation Safety Reviewer,Performance/QA). - OBSERVED — what was found, not what was done. Cite the MCP tool if relevant.
- CHANGED — each modified item with its project path. If nothing changed:
CHANGED: (none — read-only pass). - NOT CHANGED — explicit list. Always include the off-limits items confirmed untouched (stereo rig, scenario roots, slot anchors, off-limits scripts).
- RISKS — at least one item, even if
none observed. If uncertain:uncertain — verify by re-screenshot after the next change. - NEXT SAFE OPTIONS — one sentence per option, max three. Each must be a bounded change, not a refactor.
Examples
Read-only inspection
ROLE: Scene Director
OBSERVED: Active scene: LaparoscopicSurgeryEmulation. Render pipeline: URP-Balanced.asset. Scenario_01_CleanTissue is the only active scenario root. CavitySlot and AnatomySlot are empty placeholders. Console: 0 errors, 1 warning (missing reference on a stale Blending child — predates this session).
CHANGED: (none — read-only pass)
NOT CHANGED: stereo rig, Blending, UI, WebRTC, RPC objects, scenario roots, slot anchors, off-limits scripts.
RISKS: The single missing-reference warning is pre-existing and is in the preserved Blending subtree. Do not "fix" it from this agent.
NEXT SAFE OPTIONS: Materials/Assets agent may add a CavityWall material instance to CavitySlot; Lighting/Rendering agent may add a Bloom + Vignette + DoF override set to the global Volume.
Bounded material tweak
ROLE: Materials/Assets
OBSERVED: Tissue_Default.mat is currently URP/Lit, base color #B66A6A, smoothness 0.35, normal scale 0.5.
CHANGED: Assets/Materials/LaparoscopicBenchmark/Tissue_Default.mat — smoothness 0.35 → 0.42, _BumpScale 0.5 → 0.6.
NOT CHANGED: Tissue_Wet.mat, Tissue_LowTexture.mat, CavityWall.mat, InstrumentProxy.mat, all scenario roots, all slot anchors, off-limits scripts.
RISKS: Slight increase in specular response may push Scenario_01 closer to "wet". Revert _Smoothness to 0.35 if Scenario_01 must stay drier.
NEXT SAFE OPTIONS: Lighting/Rendering agent may add a tight Volume override (Vignette.intensity 0.3, LensDistortion.intensity 0.1); Performance/QA agent may re-screenshot and re-check console.
Destructive change refused
ROLE: Unity Execution
OBSERVED: User request was to delete Scenario_04_PartialOcclusion. The contract names all five scenario roots as the external-automation surface.
CHANGED: (none — destructive change refused)
NOT CHANGED: Scenario_04_PartialOcclusion (kept), all other scenario roots, slot anchors, off-limits scripts, stereo rig, Blending, UI, WebRTC, RPC.
RISKS: Deleting any Scenario_0X root breaks the external-automation contract.
NEXT SAFE OPTIONS: The scenario can be deactivated (SetActive false) instead, preserving the contract.
Cross-agent passing
Agent B's first action after receiving A's handoff: one targeted read (e.g., find_gameobjects on the path A mentioned) to verify A's claimed state is still true. If A's state has drifted, report the drift in OBSERVED and proceed cautiously.
Anti-patterns to refuse in handoffs
- Vague
OBSERVED("the scene looked OK"). CHANGEDwithout paths.NOT CHANGEDleft empty when off-limits items exist.RISKSmissing or "all good" with no evidence.NEXT SAFE OPTIONSthat are refactors instead of bounded changes.- Hiding a refused change behind a "completed" framing.
- Claiming clinical accuracy, surgical correctness, or patient-safety implications.