Index of every page on this site: llms.txt
All skills
microsoft avatar

/microsoft-foundry

@0d60658
by microsoftmicrosoft/skills3.2k stars
373
Run it yourself

Build, deploy, evaluate, optimize, fine-tune, and manage Microsoft Foundry agents, models, and resources end to end. USE FOR: foundry, azd ai agent, azd provision/deploy, hosted agent scaffold/develop/run/deploy/troubleshoot, prompt agent, managed harness agent, Copilot harness, voice agent, create/update/invoke/evaluate agent, add tool to agent, agent insights, batch/continuous eval, continuous monitoring, agent CI/CD, optimize/improve prompt, optimize agent instructions, Agent Optimizer, dataset curation from traces, deploy model, model fine-tuning (SFT/DPO/RFT), Foundry project, RBAC, role assignment, permissions, quota, capacity, region, deployment failure, AI Services, create Foundry resource, knowledge index, customize deployment, onboard, availability, training-data, grader, distillation, large file upload. DO NOT USE FOR: Azure Functions, App Service, general Azure deploy (use azure-deploy), general Azure prep (use azure-prepare).

Use this Skill: https://skilld.dev/gh/microsoft/skills/microsoft-foundry

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

foundry-agentcreatereferencesvoice-agentprompt-and-hosted-voice-settings.md

≈1.9k tokens on demand. Your agent reads this file only when SKILL.md points to it.

Voice Agent YAML Settings

For requested/preserved Prompt Voice and Hosted Voice settings. Use kind: voice; preserve unrelated fields. Do not apply this reference to GPT-Live Voice; use GPT-Live Voice YAML, whose native Live authoring surface is intentionally separate.

Owner Settings
Prompt Voice Model, instructions, inputs, audio, tools, behavior, data, telephony
Hosted wrapper Audio/voice, structuredInputs, greeting, outputModalities/avatar, store, parallelToolCalls, maxOutputTokens, include
Hosted target Model, instructions, tools, logic

Hosted wrappers must omit target/runtime-owned model, modelType, instructions, tools, toolChoice, policies, protocols, codeConfiguration, sessionConfiguration, and environment fields.

Advanced Settings

structuredInputs: { persona: { description: Persona, defaultValue: Ada } }

audio:
  input:
    format: { type: audio/pcm, rate: 24000 }
    noiseReduction: { type: near_field }
    echoCancellation: { type: server_echo_cancellation, channels: 1 }
    transcription: { model: azure-speech, language: en-US, prompt: Terms }
    turnDetection: { type: server_vad, threshold: 0.6, prefixPaddingMs: 300,
      silenceDurationMs: 500, createResponse: true, eagerness: <value>,
      speechDurationMs: 120, removeFillerWords: true, interruptResponse: true,
      languages: [en-US], autoTruncate: true }
  output:
    format: { type: audio/pcm, rate: 24000 }
    voice: { type: azure-standard, name: <voice-name>,
             style: cheerful, pitch: "+0Hz", locale: en-US, volume: "+0%" }
    speed: 1.1

defaultValue maps to wire default_value. Defaults: PCM/24000, server_vad, azure-speech, [audio], azd voice.

Known Authoring Values and Constraints

Setting Values / constraints
audio.*.format audio/pcm requires rate: 24000; audio/pcmu and audio/pcma must omit rate.
audio.input.turnDetection Known type values: server_vad, semantic_vad, azure_semantic_vad (schema is open, not an enum). threshold: >0 and <=1. prefixPaddingMs, silenceDurationMs, speechDurationMs: nonnegative integers. createResponse, removeFillerWords, interruptResponse, autoTruncate: booleans. languages: string array. Known service eagerness: auto, low, medium, high (schema is open).
audio.input.noiseReduction Nonblank type; known values: near_field, far_field, azure_deep_noise_suppression (schema is open).
audio.input.transcription model, language, prompt: strings. Known phrase-compatible models: azure-speech, azure-fast-transcription; include: [item.input_audio_transcription.phrases] requires one. Do not author phraseList; azd doesn't support it.
audio.output.voice Nested shape requires nonblank type and name; source-known types: azure-standard / azure_standard, openai. Nonempty legacy voice.rate is rejected; use audio.output.speed.
audio.output.speed 0.25-1.5.
Responses outputModalities must be nonempty; known values: audio, text, animation, avatar. maxOutputTokens: integer 1-2147483647 or inf. parallelToolCalls is a boolean for Prompt Voice and Hosted Voice wrappers. GPT-Live Voice root definitions reject it; GPT-Live Responses delegation is documented separately.
Telephony Prompt Voice providers: twilio, acs; each binding requires provider, identifier, connection. Twilio identifier: +<E.164>; ACS/TPE identifier: 28:orgid:<guid>.

Behavior and Responses

greeting: { type: template, text: "Hello {{persona}}" }
outputModalities: [audio, text]
avatar: { type: video_avatar, character: lisa, output_protocol: webrtc }
tools: [{ type: system, name: end_conversation }]
toolChoice: auto
parallelToolCalls: true
maxOutputTokens: inf # or integer 1-2147483647
include: [item.input_audio_transcription.phrases]

Preserve opaque maps.

Data Persistence

store defaults false; true persists transcript, events, audio. Preserve explicit values.

Telephony Bindings

Twilio: If a connection is supplied, show/list without credentials and verify kind Twilio, auth ApiKey, and known SID target. Otherwise resolve name (default telephony-twilio), SID, token availability, and owned voice-capable E.164 number; create before deployment:

AZURE_DEV_USER_AGENT=microsoft_foundry_skill azd ai connection create "$CONNECTION_NAME" \
  --kind Twilio --target "$TWILIO_ACCOUNT_SID" --auth-type api-key \
  --key "$TWILIO_AUTH_TOKEN" --project-endpoint "$AZURE_AI_PROJECT_ENDPOINT" --no-prompt
$env:AZURE_DEV_USER_AGENT = "microsoft_foundry_skill"
azd ai connection create $ConnectionName `
  --kind Twilio --target $env:TWILIO_ACCOUNT_SID --auth-type api-key `
  --key $env:TWILIO_AUTH_TOKEN --project-endpoint $env:AZURE_AI_PROJECT_ENDPOINT --no-prompt

Arguments can be locally visible; use the portal if concerned. Verify via connection show ... --output json, never --show-credentials. Bind:

telephony:
  bindings:
    - provider: twilio
      identifier: "+14255550123"
      connection: telephony-twilio
      transferTargets:
        - name: support
          description: Transfer the caller to support.
          destination: { kind: pstn, value: "+14255550124" }

Token: process environment only, never YAML/chat/summaries. Connection must preexist deploy. azd doesn't create accounts, buy/discover numbers, or prove routing; test a call.

Bindings require provider, identifier, and connection; transferTargets is optional. Omit it to leave the existing singleton transfer targets unmanaged, or set transferTargets: [] to manage the set as empty. When present, each item requires name, a nonblank description, and destination. Destination kinds:

  • pstn: value is +<E.164>.
  • teams: value is a canonical GUID or 28:orgid:<guid>.
  • sip: value is a valid sip: or sips: URI with user@host.

Names in the effective target set must be unique case-insensitively. Descriptions are limited to 1024 UTF-16 code units; destination values, including SIP URIs, are limited to 2048.

Teams Phone extensibility (TPE) / ACS

For an explicit Teams Phone extensibility (TPE) or ACS request, reuse a supplied TPE/ACS connection only after checking category AzureCommunicationServices, ACS endpoint, and available metadata.ResourceId. The azd compatibility authoring value is provider: acs; bind a TPE resource account:

telephony:
  bindings:
    - provider: acs
      identifier: "28:orgid:<resource-account-object-id>"
      connection: <acs-connection-name>

Legacy ACS purchased-number identifiers such as 4:+<E.164> are rejected by authoring validation. If no compatible TPE/ACS connection exists, don't create it with azd or automate RBAC/Event Grid/TPE setup. Use the guide and Foundry portal → Voice Agent → Channels → Phone numbers → Add a number → Microsoft Teams; the portal coordinates the ACS connection, project identity, TPE resource account, secure webhook/Event Grid, and TPE telephony binding. Test a real call; WebSocket smoke doesn't validate telephony.

Source: SKILL.md on GitHub

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

3 warnings7d4 checks · Risk SAFE
  • Gen Agent Trust Hub7d

    This skill provides a comprehensive toolset for managing Microsoft AI Foundry resources, authored by Microsoft. It includes some security considerations such as the dynamic execution of user-provided Python scripts for model grading and the automated execution of Azure CLI commands. While these patterns warrant standard development review, they are used appropriately within the skill's intended functionality for developer workflows and cloud resource management. See the detailed analysis for additional context.

  • Socket7d

    2 alerts: gptSecurity, gptAnomaly

  • Snyk7d

    Risk: MEDIUM · 1 issue

  • Runlayer9mo

    36/36 files flagged

Signed by skilld at 0d60658. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub yesterday.

Activeupdated 5 days ago
metadata
{
  "author": "Microsoft",
  "version": "1.2.36"
}

README badge

README badge for microsoft/skills/microsoft-foundry