원클릭으로
neuro-vrchat
Bridge Neuro API actions into safe VRChat autonomy.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
메뉴
Bridge Neuro API actions into safe VRChat autonomy.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Analyze system architecture, design patterns, module boundaries, and scalability. Provide high-level design recommendations and architectural decision records (ADR).
Blender CAD modeling automation skill with STEP/IGES import, Geometry Nodes, and USD export support.
Advanced build management agent that provides high-performance incremental compilation, hot reload capabilities, and intelligent build orchestration. Features MD5-based change detection, process management, and multi-platform deployment support with tqdm progress visualization.
Diagnose LINE and Telegram credential readiness versus live roundtrip readiness through Hypura Harness without exposing tokens or raw routing ids.
Comprehensive CI/CD integration agent that manages continuous integration and deployment pipelines with automated quality assurance, multi-platform support, and intelligent workflow orchestration. Seamlessly integrates with GitHub Actions, GitLab CI, Jenkins, CircleCI, and Azure DevOps for end-to-end DevOps automation.
Advanced code analysis agent that performs comprehensive code reviews focusing on security, performance, maintainability, and best practices. Generates actionable feedback with severity levels and automated fix suggestions.
| name | neuro-vrchat |
| description | Bridge Neuro API actions into safe VRChat autonomy. |
| version | 1.0.0 |
| author | Bob Nyan, Hermes Agent |
| license | MIT |
| platforms | ["linux","macos","windows"] |
| metadata | {"hermes":{"tags":["vrchat","neuro-api","osc","voicevox","autonomy","gaming"],"category":"gaming","related_skills":["vrchat-osc"]}} |
Use this skill when a user wants Hermes to expose a Neuro API-compatible bridge for VRChat autonomy.
vendor/neuro-sdk clone from https://github.com/VedalAI/neuro-sdk.python-osc available through the vrchat extra.http://127.0.0.1:50021 when speech is enabled.websockets==15.0.1 when running scripts/vrchat_neuro_bridge.py.vrchat_neuro_status before attempting bridge work.vrchat_neuro_build_messages to prepare startup, optional context, and actions/register messages.vrchat_neuro_handle_action for each incoming Neuro action message.vrchat_observation_ingest for STT, vision, stream, operator, or system observations.vrchat_observation_from_osc for incoming VRChat OSC ChatBox events.vrchat_observation_queue_status before a heartbeat or scheduler consumes queued context.dry_run: true until the operator intentionally tests in a private VRChat instance.vrchat_autonomy_prepare_profile to create the local private-test dry-run profile before heartbeat work.vrchat_autonomy_profile_status and vrchat_autonomy_profile_tick for profile-driven heartbeat or scheduler checks.vrchat_autonomy_heartbeat_tick when a scheduler should turn a ready launch/readiness event into one profile-driven tick.vrchat_autonomy_conversation_dry_run to prove vision, STT, ChatBox, and operator observations can plan ChatBox/VOICEVOX output and route through Neuro without live actuation.vrchat_autonomy_preflight_bundle before any private-instance live test, and inspect vrchat_process.phase, voicevox.process.phase, voicevox_synthesis, and audio.virtual_cable_route when the operator says VRChat or VOICEVOX was started but readiness is still blocked.vrchat_autonomy_runtime_doctor when operator-reported runtime state disagrees with read-only readiness; inspect its VOICEVOX URL probes, port snapshot, visible-window evidence, process visibility, and bounded launch discovery before asking the operator to retry.vrchat_autonomy_wait_ready when the operator is actively launching VRChat or VOICEVOX and wants a bounded read-only wait loop.vrchat_autonomy_wait_then_tick when the operator wants readiness wait followed by one gated profile tick.vrchat_autonomy_prepare_private_smoke immediately before any live private smoke to evaluate live gates and build a dry-run plan without live output.vrchat_autonomy_wait_then_private_smoke when the operator has launched VRChat and VOICEVOX and wants the harness to wait until readiness, then stop at preparation by default.vrchat_autonomy_private_smoke before any private-instance live test.vrchat_autonomy_completion_audit before claiming the full VRChat autonomy objective is complete.The Neuro harness is scripts/vrchat_neuro_bridge.py. It opens a websocket to a Neuro API server, sends bootstrap messages, receives action commands, routes them through vrchat_neuro_handle_action, and returns action/result.
The harness does not bypass the Hermes profile. If the profile is missing, invalid, disabled, or dry-run, no live VRChat OSC or VOICEVOX audio should be produced.
The observation harness is scripts/vrchat_observation_harness.py. It queues JSONL events from STT, vision, stream chat, or operator panels, and can also listen for incoming /chatbox/input OSC events. Its --tick-profile mode refuses live profiles unless --allow-live-profile is supplied.
The heartbeat tick harness is scripts/vrchat_heartbeat_tick.py. It first runs the read-only heartbeat, then runs one profile tick only on VRCHAT_LAUNCHED_READY or READINESS_COMPLETE unless the operator asks for --tick-when-already-ready or --force-tick. It refuses non-dry-run profiles unless --allow-live-profile and the exact live acknowledgement are both supplied.
The conversation dry-run harness is scripts/vrchat_conversation_dry_run.py. It runs representative multimodal observations through local planning and the Neuro action path, defaults to non-persistent observations, and must keep all actuation safety flags false.
The profile harness is scripts/vrchat_profile.py. It prepares an enabled private-test dry-run profile by default, with VOICEVOX and ChatBox allowed, movement blocked, the virtual cable playback side set to CABLE Input, and the VRChat microphone side set to CABLE Output. Its --arm-live mode refuses to write unless the exact live acknowledgement is supplied.
The preflight harness is scripts/vrchat_preflight.py. It collects profile, readiness, Neuro SDK vendor, observation queue, audio output device evidence, the virtual cable playback/microphone-side route, and optional no-playback VOICEVOX synthesis without sending OSC, playing audio, recording microphone input, or opening a Neuro websocket.
The runtime doctor harness is scripts/vrchat_runtime_doctor.py. It extends preflight with operator mismatch flags, common VOICEVOX local URL probes, local port snapshots, relevant visible Windows desktop windows, bounded process visibility diagnostics without storing command lines, bounded read-only launch discovery for Steam VRChat and common VOICEVOX locations, and concrete next actions. It is read-only, does not launch apps, and must keep all live actuation flags false.
The private smoke preparation harness is scripts/vrchat_private_smoke.py --prepare-only. It validates readiness, profile state, and live acknowledgement, then builds a dry-run ChatBox/VOICEVOX/avatar action plan. It must never send ChatBox, play audio, or write avatar parameters.
The wait harness is scripts/vrchat_wait_ready.py. It repeatedly collects the same read-only preflight evidence until readiness succeeds or timeout expires. It records bounded snapshots and never runs a profile tick by itself.
The wait-then-tick harness is scripts/vrchat_wait_then_tick.py. It waits for readiness, then calls the heartbeat tick path with tick_when_already_ready. It keeps the profile gate intact; non-dry-run profiles still require --allow-live-profile and the exact live acknowledgement.
The wait-then-private-smoke harness is scripts/vrchat_wait_then_private_smoke.py. It waits for read-only readiness, then runs the private smoke preparation path. It must not run live output unless --allow-live-smoke, complete readiness, a valid armed non-dry-run profile, and the exact live acknowledgement are all present.
The private smoke harness is scripts/vrchat_private_smoke.py. It defaults to dry-run and requires readiness, a valid enabled non-dry-run profile, and the exact live acknowledgement before any ChatBox, VOICEVOX, or avatar action execution.
The completion audit harness is scripts/vrchat_completion_audit.py. It is read-only and reports whether the current workspace/runtime evidence satisfies the full objective. It verifies a dry-run multimodal turn plan from synthetic observation context, no-playback VOICEVOX synthesis, and one synthetic Neuro action through the Hermes safety gate without live output. It must leave live smoke incomplete until the local runtime is ready and the operator deliberately arms live output.
Register only these Hermes-backed actions:
vrchat_autonomy_turnvrchat_speakvrchat_chatboxvrchat_avatar_action when the profile has allowed avatar actionsAvatar actions must come from profile action IDs. Do not expose OSC addresses, OSC args, VRChat credentials, client modification paths, or arbitrary code execution as Neuro actions.
For local validation, run focused compile, unit tests, registry discovery, and a read-only readiness check. A manual live smoke test requires VRChat, VOICEVOX, OSC enabled, the virtual cable selected in VRChat, and an explicit non-dry-run profile acknowledgement.