Routes 3D ResNets PyTorch video action-recognition workflows across training, inference, and data preparation.
日本語の概要は準備中です。原文の説明を表示しています。
Operate the Bindu agent microservice framework: Python bindufy agents, A2A protocol, security, gRPC SDKs, deployment, Gateway, and Inbox.
インストールする前に、エージェントに与えられる指示の中身を確認できます。
Use this skill when a task is about building, running, integrating, or troubleshooting Bindu: a Python-first framework that turns AI handlers into DID-identified A2A-speaking microservices, with optional Hydra auth, DID signatures, mTLS, x402 payments, gRPC language SDKs, boxd runtime deployment, a Gateway planner, and an operator Inbox UI.
Install the public package in the target environment when needed:
python -m pip install bindu
When a user is working with an installed Bindu package, first confirm the package and CLI surface without starting a server:
python scripts/check_bindu_install.py --json
If the user is editing a Bindu checkout, run project-native commands only after reading the relevant sub-skill and respecting repo maintenance rules in references/repo-maintenance.md.
| User intent | Read |
|---|---|
Write a Python agent, call bindufy(), validate config/handler shape, send A2A JSON-RPC, poll tasks, expose public skills, or interpret negotiation | sub-skills/agent-authoring-and-a2a/SKILL.md |
| Configure or debug Hydra OAuth, DID keys/signatures, private skill catalogs, mTLS, x402 payments, payment sessions, or auth/payment errors | sub-skills/security-identity-and-payments/SKILL.md |
Register TypeScript or other language agents, debug core :3774, callback gRPC, proto drift, generated stubs, or SDK heartbeat/handler responses | sub-skills/grpc-and-language-sdks/SKILL.md |
Use bindu serve, bindu deploy, runtime-boxd, source packaging, secret exclusion, storage/scheduler, migrations, observability, tunneling, or maintainer commands | sub-skills/deployment-runtime-and-operations/SKILL.md |
Operate Bindu Gateway /plan, recipes, peer catalogs/auth, send-and-poll orchestration, Inbox UI, personal agent, contacts, demo peers, or webhooks | sub-skills/gateway-inbox-and-orchestration/SKILL.md |
references/capability-map.md maps Bindu features to sub-skills, bundled references, and verification evidence.references/troubleshooting.md covers cross-cutting install/import, generated-code, port, optional-service, and routing problems.references/repo-maintenance.md summarizes repository contribution rules, native commands, and generated-code boundaries.references/repo-provenance.md records the source snapshot and evidence paths used to create this skill.references/repo-routing-metadata.json is structured router metadata for managed repo-skill import tooling.scripts/check_bindu_install.py verifies the installed distribution, important imports, CLI entry point, and optional environment signals. It does not start servers, contact networks, read secrets, or require the original repository checkout.proto/agent_handler.proto in a checkout and regenerate with the documented commands..env, key, wallet, cert, or token files. Use .env.example with allowlist pragmas only when the repo's secret scanner requires it.app_settings for Bindu config and get_logger(__name__) for logging when editing the repository.まだレビューはありません。使ってみた感想をお寄せください。
概要と使いどころ
Routes 3D ResNets PyTorch video action-recognition workflows across training, inference, and data preparation.
日本語の概要は準備中です。原文の説明を表示しています。
Guide 3DDFA Python inference, geometry rendering, training/evaluation, and optional C++ ONNX workflows for 3D dense face alignment.
日本語の概要は準備中です。原文の説明を表示しています。
Routes 3DDFA_V2 face-alignment setup, still-image demos, video tracking, and ONNX benchmarking workflows.
日本語の概要は準備中です。原文の説明を表示しています。
Operate AB3DMOT 3D multi-object tracking workflows for KITTI and nuScenes data, tracking, evaluation, and visualization.
日本語の概要は準備中です。原文の説明を表示しています。
Use Hugging Face Accelerate for PyTorch training-loop migration, distributed launch/configuration, DeepSpeed/FSDP/TPU backend setup, big-model inference/offload, checkpointing, tracking, and troubleshooting.
日本語の概要は準備中です。原文の説明を表示しています。
Route Acme reinforcement-learning framework tasks across core loops, replay/data, JAX agents, and TensorFlow/Sonnet agents.
日本語の概要は準備中です。原文の説明を表示しています。