| name | agentframework-repo |
| description | Repo-specific guidance for implementing and modifying the agentframework/ project in azure-ai-apps. Covers existing ChatAgent patterns, config-driven agent selection, Kaito/Azure Foundry backends, diagnostics integration, SQLite vs Azure SQL handling, and targeted validation commands. WHEN: change agentframework, add agent to chatbot, add BaseChatClient backend, modify chatbot/config.yaml, update prompts, add diagnostics, change thread or memory handling, work on Kaito integration, work on Azure SQL fitness memory, update Streamlit chatbot behavior. |
Agent Framework Repo Specialization
Use this skill for work inside agentframework/ when the task is not just about the generic Microsoft Agent Framework SDK, but about how this repository applies it.
Use This Skill For
- Adding or modifying agents in
agentframework/
- Extending
BaseChatClient implementations such as KAITO-backed clients
- Updating
chatbot/config.yaml, prompt templates, or provider selection behavior
- Working on diagnostics, thread persistence, or durable fitness memory
- Adjusting Streamlit chatbot behavior without redesigning the UI
Repo Biases
- Default to
agentframework/ when the user asks about agent orchestration in this repo.
- Preserve the current config-driven architecture instead of hardcoding providers, models, or prompts.
- Reuse existing Agent Framework SDK patterns already present in the repo before introducing new abstractions.
- Prefer targeted changes over framework-wide refactors.
Key Files
agentframework/chatbot/chatbot.py: Streamlit UI orchestration and agent interaction flow
agentframework/chatbot/config.yaml: provider and agent selection configuration
agentframework/ai_chat_client.py: custom chat client implementations, including KAITO/OpenAI-compatible backends
agentframework/main.py: agent factory patterns and assembly
agentframework/fitness_memory.py: context providers and durable memory integration
agentframework/db.py: database backend access and repository selection
agentframework/run_utils.py: retry and streaming helpers
agentframework/prompts/: prompt templates and structured prompt assets
Implementation Rules
- Keep the boundary between
ChatAgent orchestration and backend-specific client logic intact.
- Prefer changes in config, prompt files, or agent factory setup before changing the Streamlit app flow.
- When adding a new backend, follow the existing
BaseChatClient extension pattern instead of introducing a separate orchestration path.
- When adding memory or profile features, verify whether the active backend is SQLite or Azure SQL and avoid mixing assumptions.
- Reuse the existing diagnostics path and store patterns rather than inventing parallel logging or tracing systems.
Validation Defaults
- From
agentframework/, prefer uv sync and uv run ... commands.
- Use the smallest relevant validation step for the edit:
uv run ruff check .
uv run mypy .
uv run pytest
uv run streamlit run chatbot/chatbot.py
- If the change is local to one module, prefer targeted validation over full-project runs.
Azure And AKS Safety
- Read-only Azure investigation is acceptable when needed for diagnostics.
- Do not modify live Azure resources, deploy to Azure, or change AKS state unless the user explicitly asks.
- For local code changes that affect AKS deployment assets, update manifests or scripts only in the repo unless the user explicitly requests rollout actions.
Relationship To The Generic Skill
The user-level microsoft-agent-framework skill should remain the source for broad SDK guidance.
This repo skill narrows that guidance to the conventions and file structure used by azure-ai-apps/agentframework.