|
1 | 1 | # Release History |
2 | 2 |
|
3 | | -## 1.0.0-beta.5 (Unreleased) |
| 3 | +## 1.0.0-beta.5 (2026-07-09) |
4 | 4 |
|
5 | 5 | ### Features Added |
6 | 6 |
|
7 | | -- [[#8478]](https://github.com/Azure/azure-dev/issues/8478) Add `--client-header` to `azd ai agent invoke` for sending custom `x-client-*` request headers in `"Name: Value"` format (repeatable). The responses and invocations protocols forward the `x-client-*` header family to the agent; other header names are rejected, and the flag is not supported with the `a2a` protocol (which does not propagate `x-client-*` headers). Managed headers (`Authorization`, `Content-Type`, user identity) always take precedence. |
| 7 | +- [[#9043]](https://github.com/Azure/azure-dev/pull/9043) Add `--client-header` to `azd ai agent invoke` for sending custom `x-client-*` request headers in `"Name: Value"` format (repeatable). The responses and invocations protocols forward the `x-client-*` header family to the agent; other header names are rejected, and the flag is not supported with the `a2a` protocol (which does not propagate `x-client-*` headers). Managed headers (`Authorization`, `Content-Type`, user identity) always take precedence. |
8 | 8 | - [[#8939]](https://github.com/Azure/azure-dev/pull/8939) Add native support for the Activity protocol to `azd ai agent`. `azd ai agent init` can now scaffold an Activity-protocol agent (defaulting to the service-recommended version `2.0.0`), `azd deploy` provisions a companion Azure Bot Service registration authenticated with `BotServiceRbac` and prints Microsoft Teams setup guidance, and `azd down` tears the bot down. Both init-from-code and init-from-manifest flows are supported. |
| 9 | +- [[#8983]](https://github.com/Azure/azure-dev/pull/8983) `azd ai agent run` now supports activity-protocol agents in a pure-local inner loop — no Foundry deploy, no Azure Bot, no Teams sideload required. An M365 Agents Playground integration is available for local testing. Thanks @v1212 for the contribution! |
9 | 10 | - [[#8989]](https://github.com/Azure/azure-dev/pull/8989) Add `a2a` protocol support to `azd ai agent invoke`. A plain message is wrapped in a JSON-RPC 2.0 `message/send` request, `--input-file` sends a complete JSON-RPC request, and `--output raw` dumps the response verbatim. A2A is remote-only (not available with `--local`). |
| 11 | +- [[#9003]](https://github.com/Azure/azure-dev/pull/9003) Improve `azd ai agent optimize` with live-updating candidate rows, phase-aware progress indicators, azure.yaml inline/config agent detection, and `--output json` support for `optimize status`. Thanks @Zyysurely for the contribution! |
10 | 12 |
|
11 | 13 | ### Bugs Fixed |
12 | 14 |
|
13 | | -- [[#8954]](https://github.com/Azure/azure-dev/issues/8954) Fix `azd ai agent run` failing with `Python 3.13+ is required` when a compatible Python is installed but not first on `PATH`. When falling back to `pip`, azd now probes multiple interpreters (including the Windows `py -3` launcher, which selects the newest installed Python 3) and checks each one's version, selecting the first that satisfies the runtime instead of hard-failing on whichever `python` appears first on `PATH`. |
14 | | -- [[#8411]](https://github.com/Azure/azure-dev/issues/8411) Fix `azd ai agent invoke --local` failing with `could not connect to localhost:<port>` when run immediately after `azd ai agent run`, because the agent's listener binds a few seconds after `run` starts. `invoke --local` now retries connection-refused errors with backoff for up to 60s, and `azd ai agent run` waits up to 90s (covering slow interpreter startup and agent-stack imports before the server binds) for the port to accept connections before printing its "Agent ready" signal. |
15 | | -- [[#8211]](https://github.com/Azure/azure-dev/issues/8211) Update RBAC callouts in developer role checks and `azd ai doctor` to use the renamed Foundry built-in role names (`Foundry User`, `Foundry Project Manager`, `Foundry Account Owner`; formerly `Azure AI User`/`Project Manager`/`Account Owner`). The suggested `az role assignment create` commands now reference the role by its GUID so they keep working regardless of the display-name rollout. |
16 | | -- [[#8839]](https://github.com/Azure/azure-dev/issues/8839) Fix `azd down` on a Foundry (`microsoft.foundry`) project failing outright without `--force`. It now prompts for confirmation (naming the resource group to be deleted, defaulting to "no") like the built-in Bicep provider, and only falls back to requiring `--force` when there is no interactive terminal (for example under `--no-prompt` or in CI). |
| 15 | +- [[#9012]](https://github.com/Azure/azure-dev/pull/9012) Fix `azd ai agent run` failing with `Python 3.13+ is required` when a compatible Python is installed but not first on `PATH`. When falling back to `pip`, azd now probes multiple interpreters (including the Windows `py -3` launcher, which selects the newest installed Python 3) and checks each one's version, selecting the first that satisfies the runtime instead of hard-failing on whichever `python` appears first on `PATH`. |
| 16 | +- [[#9044]](https://github.com/Azure/azure-dev/pull/9044) Fix `azd ai agent invoke --local` failing with `could not connect to localhost:<port>` when run immediately after `azd ai agent run`, because the agent's listener binds a few seconds after `run` starts. `invoke --local` now retries connection-refused errors with backoff for up to 60s, and `azd ai agent run` waits up to 90s (covering slow interpreter startup and agent-stack imports before the server binds) for the port to accept connections before printing its "Agent ready" signal. |
| 17 | +- [[#9041]](https://github.com/Azure/azure-dev/pull/9041) Update RBAC callouts in developer role checks and `azd ai doctor` to use the renamed Foundry built-in role names (`Foundry User`, `Foundry Project Manager`, `Foundry Account Owner`; formerly `Azure AI User`/`Project Manager`/`Account Owner`). The suggested `az role assignment create` commands now reference the role by its GUID so they keep working regardless of the display-name rollout. |
| 18 | +- [[#9022]](https://github.com/Azure/azure-dev/pull/9022) Fix `azd down` on a Foundry (`microsoft.foundry`) project failing outright without `--force`. It now prompts for confirmation (naming the resource group to be deleted, defaulting to "no") like the built-in Bicep provider, and only falls back to requiring `--force` when there is no interactive terminal (for example under `--no-prompt` or in CI). |
17 | 19 | - [[#8987]](https://github.com/Azure/azure-dev/pull/8987) Fix `azd ai agent init -m <manifest>` not prompting for the agent name. The prompt default and project folder are now derived from the manifest's `template.name` (falling back to the top-level `name`), matching the interactive and template flows. |
18 | 20 | - [[#8981]](https://github.com/Azure/azure-dev/pull/8981) Fix `azd ai agent init -m <azure.yaml> --deploy-mode container` not resolving a container registry when adopting a unified Foundry `azure.yaml` on an existing Foundry project, which made `azd deploy` fail with `could not determine container registry endpoint`. The deploy mode is now resolved before Foundry project setup, so a container agent wires `AZURE_CONTAINER_REGISTRY_ENDPOINT` (or is signaled to create one on provision) while code deploy and `--image` still skip ACR. |
| 21 | +- [[#9051]](https://github.com/Azure/azure-dev/pull/9051) Fix `azd ai agent init` with "Use an existing Foundry project" not stamping the `endpoint:` field on the `azure.ai.project` service in `azure.yaml`, which caused `azd up` to provision a brand-new account/project instead of reusing the selected one. |
| 22 | + |
| 23 | +### Other Changes |
| 24 | + |
| 25 | +- [[#8866]](https://github.com/Azure/azure-dev/pull/8866) Remove `Foundry-Features: *=V1Preview` opt-in headers now that Foundry hosted-agents, code-agents, and toolbox APIs are GA. |
19 | 26 |
|
20 | 27 | ## 1.0.0-beta.4 (2026-07-03) |
21 | 28 |
|
|
0 commit comments