Skip to main content

Copilot Artifacts

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

TypeNameDescriptionPath
AgentDataviewer DeveloperInteractive dataset analysis and tool development.github/agents/dataviewer-developer.agent.md
AgentOSMO Training ManagerLeRobot training lifecycle on OSMO with Azure ML.github/agents/osmo-training-manager.agent.md
InstructionCommit MessagesConventional Commits format for all commit messages.github/instructions/commit-message.instructions.md
InstructionDataviewerCoding standards for dataviewer development.github/instructions/dataviewer.instructions.md
InstructionDocs Style and ConventionsWriting standards for all markdown files.github/instructions/docs-style-and-conventions.instructions.md
InstructionRPI TrackingShared RPI working-artifact conventions.github/instructions/hve-core/copilot-tracking.instructions.md
InstructionShell ScriptsImplementation standards for bash scripts.github/instructions/shell-scripts.instructions.md
Prompt/chatlogCreate and maintain conversation logs.github/prompts/chatlog.prompt.md
Prompt/check-training-statusMonitor OSMO training job progress.github/prompts/check-training-status.prompt.md
Prompt/start-dataviewerLaunch Dataset Analysis Tool.github/prompts/start-dataviewer.prompt.md
Prompt/submit-lerobot-trainingSubmit LeRobot training job to OSMO.github/prompts/submit-lerobot-training.prompt.md
SkilldataviewerDataset browsing, annotation, and export.github/skills/dataviewer/SKILL.md
Skillenvironment-deploymentGenerate and consume environment deployment bundles.github/skills/environment-deployment/SKILL.md
Skillfleet-deploymentDeploy trained policies through fleet GitOps.github/skills/fleet-deployment/SKILL.md
Skillfleet-intelligenceMonitor fleet telemetry and drift.github/skills/fleet-intelligence/SKILL.md
SkillinfrastructureDeploy and manage Azure infrastructure.github/skills/infrastructure/SKILL.md
Skillosmo-lerobot-trainingTraining submission, monitoring, and analysis.github/skills/osmo-lerobot-training/SKILL.md
Skillsynthetic-dataGenerate synthetic robotics training data.github/skills/synthetic-data/SKILL.md
Skillrpi-*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 prompt → Dataviewer Developer
Browse and annotate training episodesDataviewer Developer agent
Submit a LeRobot training job/submit-lerobot-training prompt → OSMO Training Manager
Check training job status/check-training-status prompt → OSMO Training Manager
Save a conversation log/chatlog prompt
Run the full RPI lifecyclerpi-quick skill
Enforce commit message standardscommit-message instruction (auto-applied)
Enforce coding standards in dataviewerdataviewer instruction (auto-applied)
Enforce markdown writing standardsdocs-style-and-conventions instruction (auto-applied)
Enforce shell script standardsshell-scripts instruction (auto-applied)

🤖 Agents

Dataviewer Developer

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

PropertyValue
HandoffsStart Dataviewer, Browse Dataset, Annotate Episodes
ToolsAll (no restrictions)
Skilldataviewer
Prompts/start-dataviewer

Four-phase workflow: Launch/Configure → Interactive Browsing (Playwright) → Episode Annotation (API+UI) → Feature Development (React+FastAPI).

OSMO Training Manager

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

PropertyValue
HandoffsSubmit Training Job, Check Training Status, Run Inference Evaluation
Tools11 explicit (run_in_terminal, memory, runSubagent, ...)
Skillosmo-lerobot-training
Prompts/submit-lerobot-training, /check-training-status

Five-phase workflow: Submit → Monitor → Analyze → Summarize → Inference Evaluation. Handles VM eviction recovery, CUDA errors, and KeyError failures.

📝 Instructions

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

NameApplies ToPurpose
Commit Messages**Conventional Commits format, scopes, line-length limits
Dataviewerdata-management/viewer/**SOLID principles, test-first, validation commands
Docs Style and Conventions**/*.mdDocument hierarchy, tables, voice/tone, frontmatter
Shell Scripts**/*.shScript template, library functions, deployment patterns

⚡ Prompts

Prompts are slash commands invoked via / in the chat input. Each prompt targets a specific agent.

CommandAgent TargetRequired Inputs
/chatlogGenericNone
/check-training-statusOSMO Training ManagerworkflowId (optional)
/start-dataviewerDataviewer DeveloperdatasetPath
/submit-lerobot-trainingOSMO Training Managerdataset (required)

🛠️ Skills

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

dataviewer

PropertyValue
Directory.github/skills/dataviewer/
Resourcesreferences/PLAYWRIGHT.md (selectors, interaction recipes, API endpoints)
Used byDataviewer Developer agent

osmo-lerobot-training

PropertyValue
Directory.github/skills/osmo-lerobot-training/
Resourcesreferences/DEFAULTS.md (env, datasets, GPU profiles), references/REFERENCE.md (CLI, inference, AzureML navigation)
Used byOSMO 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.

SkillPurpose
rpi-quickCoordinate the complete RPI lifecycle
rpi-researchGather evidence and produce planning-ready findings
rpi-planBuild and critique an implementation plan
rpi-implementExecute approved plan phases and validate changes
rpi-reviewReview implementation evidence and route follow-ups
rpi-challengerChallenge scope and assumptions
rpi-plan-critiqueAssess plans independently
rpi-walkthroughWalk through RPI artifacts and decisions

🔄 Workflow Chains

Agents compose prompts and skills into end-to-end workflows:

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

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

Standalone:
├── /chatlog (prompt, 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-prompt — Create a new prompt file
  • /create-skill — Create a new agent skill

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

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.