Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/microsoft/agent-framework/foundry-hosted-agent-validationnpx skills add microsoft/agent-framework --skill foundry-hosted-agent-validationgit clone --depth 1 https://github.com/microsoft/agent-frameworkWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00082 | $0.03452 |
| Opus 5 | $0.00041 | $0.01726 |
| Sonnet 5 | $0.00016 | $0.00690 |
| Haiku 4.5 | $0.00008 | $0.00345 |
Grade A, and why
foundry-hosted-agent-validation scanned grade A with 2 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
curl -X POST http://localhost:8088/responses -H "Content-Type: application/json" \ Runs shell commandslowCapability
Expected in a hook, worth knowing in a rule or an instructions file.
`subprocess.Popen([sys.executable, "<sample>/main.py"])` (path relative to How it starts
The opening of the file, as written. The whole thing — 317 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Validating a Foundry Hosted Agent Sample
A hosted agent sample is "validated" when it passes three independent checks, plus cleanup:
- Local, native runtime — run the sample's own entry point
(
python main.py) and invoke it over HTTP. - Local, via
azd ai agent run— theazdlocal dev loop. - Deployed —
azd deployto Foundry, then invoke the hosted agent.
Each check must succeed for single-turn and multi-turn (session /
previous_response_id) conversation. Always end with cleanup (delete the
deployed agent, remove the temp azd project, restore the sample dir).
Read the sample's own
README.mdand the parent.../foundry-hosted-agents/README.mdfirst — they define the run/deploy commands and any sample-specific payload. This skill captures the process and the non-obvious gotchas the READMEs don't.
Inputs you need before starting
Gather these:
- Foundry project endpoint, e.g.
https://<account>.services.ai.azure.com/api/projects/<project>. - Foundry project resource id (for non-interactive
azd ai agent init):/subscriptions/<sub>/resourceGroups/<rg>/providers/Microsoft.CognitiveServices/accounts/<account>/projects/<project>. Find it withaz cognitiveservices account list+ the project name. - A real, deployed model name in that project (e.g.
gpt-4.1-mini). This is often different from the model id inagent.manifest.yaml— the actual deployment name wins. - An existing ACR to reuse for deployment (login server, e.g.
myacr.azurecr.io). Reusing one avoidsazd provisioncreating resources. - Whether a like-named agent already exists in the project (remove it first for a clean validation — see below).
Tooling / auth
az(logged in:az login) andazd(logged in:azd auth login).azdagents extension:azd extension listshould showazure.ai.agents; install withazd extension install azure.ai.agents.uvfor the native-Python local run.pythonneed not be on PATH —uvandazd ai agent runprovision their own interpreter.- Docker is not required when you reuse an ACR (
remoteBuild: truebuilds in ACR Tasks).
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 317 lines · 82 tokens per session scan A 9ef486e8b046
foundry-hosted-agent-validation is a skill published in the GitHub repository microsoft/agent-framework (13,222 stars, last pushed 2d ago), licensed MIT. It adds 82 tokens to every session and 3,452 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 2 findings (makes network calls, runs shell commands). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
swarmclaw
AI agent runtime and multi-agent orchestration platform. Teaches agents how to use SwarmClaw's 6 primitive tools, persistent memory, dreaming, delegation, connectors, credentials, and the skill system. Use when an agent is running on SwarmClaw and needs to understand the platform's capabilities.
agent-collaboration
Use this skill when coordinating multiple AI agents. Covers multi-agent patterns, handoffs, and orchestration strategies.
crewai-multi-agent
Multi-agent orchestration framework for autonomous AI collaboration. Use when building teams of specialized agents working together on complex tasks, when you need role-based agent collaboration with memory, or for production workflows requiring sequential/hierarchical execution. Built without LangChain dependencies…
company-product-context
Compiles comprehensive company product context from PDF documents, web research, and industry knowledge.
deep-researcher
Performs comprehensive, multi-layered research on any topic with structured analysis and synthesis of information from multiple sources.
codebase-context-extractor
This skill provides a comprehensive context extraction system for large codebases. It intelligently analyzes code structure, dependencies, and relationships to extract relevant context for understanding, debugging, or modifying code.