Skip to content

Latest commit

 

History

History
238 lines (189 loc) · 16 KB

copilot-artifacts.md

File metadata and controls

238 lines (189 loc) · 16 KB
title Copilot Artifacts
description Inventory and reference for GitHub Copilot agents, instructions, and skills configured in this repository.
author Microsoft Robotics-AI Team
ms.date 2026-10-31
ms.topic reference
keywords
copilot
agents
instructions
skills
extensibility

GitHub Copilot extensibility artifacts provide AI-assisted workflows for dataset analysis, training job management, and coding standards enforcement. Committed artifacts activate automatically in VS Code. The cloud-agent setup workflow provisions the RPI skill suite at runtime.

📋 Artifact Inventory

Type Name Description Path
Agent Dataviewer Developer Interactive dataset analysis and tool development .github/agents/dataviewer-developer.agent.md
Agent OSMO Training Manager LeRobot training lifecycle on OSMO with Azure ML .github/agents/osmo-training-manager.agent.md
Agent Dependabot PR Reviewer Dependency update review .github/agents/dependabot-pr-reviewer.agent.md
Agent Agentic Workflows GitHub agentic workflow authoring .github/agents/agentic-workflows.agent.md
Instruction Commit Messages Conventional Commits format for all commit messages .github/instructions/commit-message.instructions.md
Instruction Dataviewer Coding standards for dataviewer development .github/instructions/dataviewer.instructions.md
Instruction Docs Style and Conventions Writing standards for all markdown files .github/instructions/docs-style-and-conventions.instructions.md
Instruction RPI Tracking Shared RPI working-artifact conventions .github/instructions/hve-core/copilot-tracking.instructions.md
Instruction Shell Scripts Implementation standards for bash scripts .github/instructions/shell-scripts.instructions.md
Skill /chatlog (workflow) Create and maintain conversation logs .github/skills/chatlog/SKILL.md
Skill /check-training-status (workflow) Monitor OSMO training job progress .github/skills/check-training-status/SKILL.md
Skill /dataviewer-annotate (workflow) Annotate dataset episodes .github/skills/dataviewer-annotate/SKILL.md
Skill /start-dataviewer (workflow) Launch Dataset Analysis Tool .github/skills/start-dataviewer/SKILL.md
Skill /submit-lerobot-training (workflow) Submit LeRobot training job to OSMO .github/skills/submit-lerobot-training/SKILL.md
Skill azureml-k3s-compute-target-setup Attach an Arc-connected K3s GPU cluster to Azure ML .github/skills/azureml-k3s-compute-target-setup/SKILL.md
Skill dataviewer Dataset browsing, annotation, and export .github/skills/dataviewer/SKILL.md
Skill e2e-infrastructure Deploy and undeploy cloud E2E infrastructure .github/skills/e2e-infrastructure/SKILL.md
Skill e2e-tests Launch and monitor cloud E2E tests .github/skills/e2e-tests/SKILL.md
Skill environment-deployment Generate and consume environment deployment bundles .github/skills/environment-deployment/SKILL.md
Skill fleet-deployment Deploy trained policies through fleet GitOps .github/skills/fleet-deployment/SKILL.md
Skill fleet-intelligence Monitor fleet telemetry and drift .github/skills/fleet-intelligence/SKILL.md
Skill infrastructure Deploy and manage Azure infrastructure .github/skills/infrastructure/SKILL.md
Skill osmo-lerobot-training Training submission, monitoring, and analysis .github/skills/osmo-lerobot-training/SKILL.md
Skill synthetic-data Generate synthetic robotics training data .github/skills/synthetic-data/SKILL.md
Skill rpi-* Cloud-agent research, plan, implement, review .github/skills/rpi-*/ (runtime-provisioned, gitignored)

🔗 Quick Reference

Want to... Use this artifact
Launch the Dataset Analysis Tool /start-dataviewer skill → Dataviewer Developer
Browse and annotate training episodes Dataviewer Developer agent
Deploy or remove cloud E2E infrastructure e2e-infrastructure skill
Run and monitor cloud E2E tests e2e-tests skill
Submit a LeRobot training job /submit-lerobot-training skill → OSMO Training Manager
Check training job status /check-training-status skill → OSMO Training Manager
Save a conversation log /chatlog skill
Run the full RPI lifecycle rpi-quick skill
Enforce commit message standards commit-message instruction (auto-applied)
Enforce coding standards in dataviewer dataviewer instruction (auto-applied)
Enforce markdown writing standards docs-style-and-conventions instruction (auto-applied)
Enforce shell script standards shell-scripts instruction (auto-applied)

🤖 Agents

Dataviewer Developer

Interactive agent for launching, browsing, annotating, and improving the Dataset Analysis Tool.

Property Value
Handoffs Start Dataviewer, Browse Dataset, Annotate Episodes
Tools All (no restrictions)
Skill dataviewer
Skills /start-dataviewer, /dataviewer-annotate

Five-phase workflow: Launch/Configure → Interactive Browsing (Playwright) → Episode Annotation (API+UI) → Feature Development (React+FastAPI) → VLM-as-Judge Evaluation.

OSMO Training Manager

Multi-turn agent for managing LeRobot imitation learning training lifecycle on OSMO with Azure ML integration.

Property Value
Handoffs Submit Training Job, Check Training Status
Tools execute, read, edit, search, agent, todo, playwright/*
Skill osmo-lerobot-training
Skills /submit-lerobot-training, /check-training-status

Five-phase workflow: Submit → Monitor → Analyze → Summarize → Inference Evaluation. Handles VM eviction recovery, CUDA errors, and KeyError failures. The tools are Copilot aliases, so the agent keeps terminal, file, search, delegation, and todo access in both the Local and Copilot harnesses. playwright/* resolves only when you configure a Playwright MCP server named playwright; see the dataviewer skill for an example.

📝 Instructions

Instructions activate automatically when files matching their applyTo pattern appear in the chat context.

Name Applies To Purpose
Commit Messages ** Conventional Commits format, scopes, line-length limits
Dataviewer data-management/viewer/** SOLID principles, test-first, validation commands
Docs Style and Conventions **/*.md Document hierarchy, tables, voice/tone, frontmatter
Shell Scripts **/*.sh Script template, library functions, deployment patterns

🛠️ Skills

Skills provide multi-file capabilities with progressive 3-level loading: discovery (frontmatter only) → instructions (SKILL.md body) → resources (bundled reference files).

Workflow skills

Workflow skills are slash commands invoked via / in the chat input. They replace the former prompt files, which the Copilot harness does not load, and keep the same command names. Each sets disable-model-invocation: true so it runs only when you type /name, never on its own. A workflow that belongs to an agent links that agent's protocol, so it also works from other agents.

Command Follows Arguments
/chatlog Any agent mode, log, maintain (all optional)
/check-training-status OSMO Training Manager workflowId, metrics, summary (all optional)
/start-dataviewer Dataviewer Developer datasetPath, ports, VLM judge options (optional)
/dataviewer-annotate Dataviewer Developer datasetId (required), judge options
/submit-lerobot-training OSMO Training Manager dataset (required), training options

dataviewer

Property Value
Directory .github/skills/dataviewer/
Resources references/PLAYWRIGHT.md (selectors, interaction recipes, API endpoints)
Used by Dataviewer Developer agent

e2e-infrastructure

Property Value
Directory .github/skills/e2e-infrastructure/
Resources scripts/manage-e2e-infrastructure.sh, templates/terraform.tfvars.example
Used by Copilot sessions managing dedicated cloud E2E infrastructure

Configure user-specific governance, naming, and subscription requirements through the skill's personal environment overrides.

e2e-tests

Property Value
Directory .github/skills/e2e-tests/
Resources scripts/run-e2e-tests.sh
Used by Copilot sessions launching and monitoring cloud E2E tests

osmo-lerobot-training

Property Value
Directory .github/skills/osmo-lerobot-training/
Resources references/DEFAULTS.md (env, datasets, GPU profiles), references/REFERENCE.md (CLI, inference, AzureML navigation)
Used by OSMO Training Manager agent

rpi-* skill suite

The cloud-agent setup workflow uses gh skill install to download eight explicit RPI skill paths from a pinned microsoft/hve-core commit into gitignored .github/skills/rpi-*/ directories. It also downloads the shared copilot-tracking.instructions.md from the same commit into .github/instructions/hve-core/.

Each skill is an immediate child of .github/skills/, as required for discovery. Local clones do not contain these runtime artifacts unless the setup workflow has provisioned them. Each skill retains its upstream references and templates, and the GitHub CLI injects the source commit and tree metadata into its SKILL.md front matter.

Skill Purpose
rpi-quick Coordinate the complete RPI lifecycle
rpi-research Gather evidence and produce planning-ready findings
rpi-plan Build and critique an implementation plan
rpi-implement Execute approved plan phases and validate changes
rpi-review Review implementation evidence and route follow-ups
rpi-challenger Challenge scope and assumptions
rpi-plan-critique Assess plans independently
rpi-walkthrough Walk through RPI artifacts and decisions

🔄 Workflow Chains

Agents compose workflow skills and capability skills into end-to-end workflows:

OSMO Training Manager (agent)
  ├── /submit-lerobot-training (skill)
  ├── /check-training-status   (skill)
  └── osmo-lerobot-training    (skill)
       ├── references/DEFAULTS.md
       └── references/REFERENCE.md

Dataviewer Developer (agent)
  ├── /start-dataviewer        (skill)
  ├── /dataviewer-annotate     (skill)
  └── dataviewer               (skill)
       └── references/PLAYWRIGHT.md

Standalone:
  ├── /chatlog                 (skill, generic)
  └── rpi-quick                (skill)
       ├── rpi-research
       ├── rpi-plan
       ├── rpi-implement
       └── rpi-review

➕ Adding New Artifacts

VS Code provides generator commands for scaffolding new artifacts:

  • /create-agent — Create a new custom agent
  • /create-instruction — Create a new instruction file
  • /create-skill — Create a new agent skill, including user-invoked workflow skills

Place new artifacts in the corresponding .github/ subdirectory and update this inventory page.

Related Documentation

For broader project context, see these companion guides:


🤖 Crafted with precision by ✨Copilot following brilliant human instruction, then carefully refined by our team of discerning human reviewers.