Python: New Foundry Hosted Agents samples: RAG, Skills, and Memory

[TaoChenOSU]

Motivation and Context

We need more samples for MAF with Foundry Hosted Agents.

Description

This PR adds three:

• Integration with AzureAISearch for RAG scenarios

• Integration with Foundry Skills

• Integration with Foundry Memory

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR follows the Contribution Guidelines
• All unit tests pass, and I have added new tests where possible
• Is this a breaking change? If yes, add "[BREAKING]" prefix to the title of the PR.

[github-actions]

Automated Code Review

Reviewers: 4 | Confidence: 62%

✓ Correctness

No actionable issues found in this dimension.

✓ Security Reliability

No actionable issues found in this dimension.

✓ Test Coverage

No actionable issues found in this dimension.

✗ Design Approach

Two design issues stand out. The Azure Search RAG sample is built and documented as if sourceName/sourceLink are injected into model context for citations, but the actual AzureAISearchContextProvider only emits document text plus doc_id, so those schema fields never influence responses. Separately, the Foundry Skills sample downloads remote ZIPs and unpacks them with extractall() without any path containment check, which bypasses the repo’s own path-traversal safeguards for file-based skills and allows a crafted archive to write outside downloaded_skills/<name>. That does not advance the sample feature itself, but it does create broad lockfile churn and merge-conflict risk in a repo that explicitly prefers targeted lock updates.

Flagged Issues

• The RAG sample's citation design relies on sourceName/sourceLink fields that the provider never reads: AzureAISearchContextProvider._extract_document_text() only pulls from content-like fields plus doc_id (see _context_provider.py:980-995 and tests at test_aisearch_context_provider.py:884-914). Either the sample needs a custom provider that formats those fields into context, or the docs/schema should be reduced to what the provider actually consumes.

---

Automated review by TaoChenOSU's agents

[Copilot]

Pull request overview

This PR adds Python Foundry-hosted Responses samples demonstrating Azure AI Search RAG and Foundry Skills, plus updates the hosted sample index to include them.

Changes:

• Adds an Azure AI Search RAG hosted-agent sample with provisioning guidance, runtime code, Docker packaging, and deployment manifests.
• Adds a Foundry Skills hosted-agent sample with sample skills, skill provisioning, startup download/bootstrap, Docker packaging, and deployment manifests.
• Updates the Foundry hosted-agent README sample list.

Reviewed changes

Copilot reviewed 21 out of 22 changed files in this pull request and generated 8 comments.

Show a summary per file

File	Description
python/samples/04-hosting/foundry-hosted-agents/README.md	Adds RAG and Skills samples to the Responses sample table.
python/samples/04-hosting/foundry-hosted-agents/responses/08_azure_search_rag/.dockerignore	Excludes local/dev files from the RAG image build.
python/samples/04-hosting/foundry-hosted-agents/responses/08_azure_search_rag/.env.example	Documents required RAG environment variables.
python/samples/04-hosting/foundry-hosted-agents/responses/08_azure_search_rag/Dockerfile	Packages the RAG sample as a Python container.
python/samples/04-hosting/foundry-hosted-agents/responses/08_azure_search_rag/README.md	Documents setup, provisioning, running, and deployment for RAG.
python/samples/04-hosting/foundry-hosted-agents/responses/08_azure_search_rag/agent.manifest.yaml	Adds Foundry deployment manifest for the RAG sample.
python/samples/04-hosting/foundry-hosted-agents/responses/08_azure_search_rag/agent.yaml	Adds hosted-agent runtime config for RAG.
python/samples/04-hosting/foundry-hosted-agents/responses/08_azure_search_rag/main.py	Implements the RAG hosted agent using Azure AI Search context.
python/samples/04-hosting/foundry-hosted-agents/responses/08_azure_search_rag/provision_index.py	Adds a helper to create/seed the Azure AI Search index.
python/samples/04-hosting/foundry-hosted-agents/responses/08_azure_search_rag/requirements.txt	Lists RAG sample dependencies.
python/samples/04-hosting/foundry-hosted-agents/responses/09_foundry_skills/.dockerignore	Excludes provisioning/source/runtime artifacts from the Skills image.
python/samples/04-hosting/foundry-hosted-agents/responses/09_foundry_skills/.env.example	Documents required Skills sample environment variables.
python/samples/04-hosting/foundry-hosted-agents/responses/09_foundry_skills/Dockerfile	Packages the Skills sample as a Python container.
python/samples/04-hosting/foundry-hosted-agents/responses/09_foundry_skills/README.md	Documents Foundry Skills authoring, provisioning, runtime loading, and deployment.
python/samples/04-hosting/foundry-hosted-agents/responses/09_foundry_skills/agent.manifest.yaml	Adds Foundry deployment manifest for the Skills sample.
python/samples/04-hosting/foundry-hosted-agents/responses/09_foundry_skills/agent.yaml	Adds hosted-agent runtime config for Skills.
python/samples/04-hosting/foundry-hosted-agents/responses/09_foundry_skills/main.py	Implements startup skill download and hosted agent wiring.
python/samples/04-hosting/foundry-hosted-agents/responses/09_foundry_skills/provision_skills.py	Adds a helper to upload local SKILL.md packages to Foundry Skills.
python/samples/04-hosting/foundry-hosted-agents/responses/09_foundry_skills/requirements.txt	Lists Skills sample dependencies.
python/samples/04-hosting/foundry-hosted-agents/responses/09_foundry_skills/skills/escalation-policy/SKILL.md	Adds escalation-policy skill content.
python/samples/04-hosting/foundry-hosted-agents/responses/09_foundry_skills/skills/support-style/SKILL.md	Adds support-style skill content.

[github-actions]

Automated Code Review

Reviewers: 4 | Confidence: 61%

✓ Correctness

Three new Foundry Hosted Agents samples (RAG, Skills, Memory) are well-structured and mostly correct. API usage matches framework signatures, the zip-slip guard is properly implemented, and previous review comments appear addressed. One correctness concern: the memory sample creates an async DefaultAzureCredential without closing it, unlike the skills sample (which uses async with) and existing samples (which use the sync credential).

✓ Security Reliability

The new samples are well-structured with good security practices (zip-slip guard in the skills sample, proper async-with credential management in provisioning scripts). The one remaining issue is the Foundry Memory sample (10_foundry_memory/main.py) where an async DefaultAzureCredential is created inline and never closed—the same pattern that was previously flagged and fixed in the other two samples in this PR.

✓ Test Coverage

No actionable issues found in this dimension.

✗ Design Approach

The new Foundry Memory sample has one design-level issue: it documents per-user cross-session memory isolation, but the implementation passes a literal {{$userId}} scope string into FoundryMemoryProvider and nothing in the hosting path shown here substitutes that placeholder. That means all callers would share the same memory scope instead of being isolated by user.

Flagged Issues

• python/samples/04-hosting/foundry-hosted-agents/responses/10_foundry_memory/main.py:46 uses scope="{{$userId}}", but FoundryMemoryProvider forwards self.scope verbatim to the memory store (_memory_provider.py:170-173,198-203,259-264) and the Responses host only passes message history into agent.run (_responses.py:296-305). As written, this collapses all users into one shared memory scope rather than isolating memories per user.

---

Automated review by TaoChenOSU's agents

[github-actions]

This sample's per-user isolation depends on {{$userId}} being replaced at runtime, but the code path never performs that substitution. FoundryMemoryProvider sends self.scope verbatim to search_memories / begin_update_memories, and the Responses host only forwards messages/history into agent.run. Every caller shares the literal scope string {{$userId}}, so memories bleed across users instead of being isolated.