Changelog

0.14.0 — 2026-09-04

Features

manage isolated Work IQ and eval CLIs

Fixes

use canonical install recovery commands

0.13.1 — 2026-09-01

Fixes

harden failure isolation and unavailable-command fallback
correct oauth/register binding and let add connector declare auth
accept Copilot local marketplace installs

0.13.0 — 2026-08-28

Features

The wiqd plugin lifecycle — create, provision, share, and delete — now runs inside wiqd itself instead of shelling out to a separate toolkit, so it starts faster and surfaces failures directly. These commands now ask for Microsoft 365 sign-in rather than a separate toolkit sign-in; use wiqd auth login --interactive.
The built-in agent backend now covers the full lifecycle command surface, so wiqd agent and wiqd plugin behave consistently whether a command runs inside wiqd or through the Microsoft 365 Agents Toolkit. wiqd doctor reports the backend’s health alongside its other environment checks.
Extensions can declare MCP servers in their manifest, and wiqd aggregates them into the composed plugin so they load automatically when your host CLI installs it — no per-server registration and no separate plugin install. wiqd ext show lists the servers an extension contributes.
wiqd agent show now runs through wiqd’s built-in agent backend, so it works without the Microsoft 365 Agents Toolkit installed. Choose the environment to inspect with --env.
Extensions can contribute Copilot skills, which wiqd composes into the plugin so they reach the host CLI without a separate install.
Telemetry from the built-in agent backend now flows through wiqd’s own consent-aware pipeline, so it honors your telemetry setting and is redacted for personal data and secrets like every other wiqd event.
Shipped agent-authoring guidance now covers attaching Graph connectors, embedding agent skills, and adding MCP actions with wiqd commands, and documents which MCP authentication modes are supported. wiqd agent add action now requires explicit METHOD /path operation selectors.
Extension manifests can declare transformFailureNonFatal, so a failed post-invocation transform degrades the render instead of failing a write command whose upstream work already succeeded; the --json envelope carries transformFailed so a consumer can tell that apart from having nothing to report. @microsoft/wiqd-extension-sdk adds a ./digest export with sha256File, sha256Fd, and readAndDigestFile for size-capped SHA-256 hashing of a file or an already-open descriptor.
wiqd agent eval adds --judge-backend and --log-level. The default github-copilot judge runs evaluations with no Azure setup, while --judge-backend azure opts into Azure OpenAI for GPT-5.x and o-series models and for custom .prompty evaluators. New reference material documents which evaluators work under each backend.
default lifecycle to wiqd core

Fixes

When wiqd update fails, it now reports the real cause — a permission problem, a network failure, or an npm error — instead of generic installer boilerplate, and keeps the actionable recovery steps alongside it. The sanitized root cause appears in both the readable output and the --json envelope.
wiqd agent add auth --auth-type now accepts an explicit set of values — bearer-token (the default), api-key, oauth, and the new microsoft-entra — and rejects anything else up front instead of failing partway through. microsoft-entra needs only --scope; it does not ask for authorization or token URLs.
A command whose write already succeeded no longer reports failure when an optional display-only transform fails afterwards, so a formatting error can no longer make completed work look like it needs retrying.
Commands are listed alphabetically in wiqd --help and in the command list shown after a parse error, so a command is easier to find.
wiqd auth login --interactive now fails immediately with a clear message when it is not attached to a real terminal, instead of hanging with no visible prompt. It exits 2 with the stable JSON error code AUTH_INTERACTIVE_REQUIRES_TTY. The default non-interactive wiqd auth login is unchanged and still works headlessly in CI.
Mocked runs of the plugin lifecycle now mirror what the real commands do, so offline previews and demos reflect actual behavior.
report measured versions during CLI updates

Performance

Running wiqd with no arguments now lists the same commands as wiqd --help, and help renders about four times faster.

0.12.2 — 2026-08-14

Maintenance release — no user-facing changes to the CLI, skills, or references.

0.12.1 — 2026-08-13

Maintenance release — no user-facing changes to the CLI, skills, or references.

0.12.0 — 2026-08-11

Features

Share an agent through a guided flow that helps you pick the target — one or more people, a mail-enabled security group, or your whole tenant — and inspects the project first so you confirm what will be shared before it happens

Fixes

wiqd plugin import and wiqd plugin export now refuse a plugin whose declarative agent references a file outside the plugin’s own package, and stop on a corrupt source manifest instead of quietly importing without the agent. wiqd plugin create now tells a non-empty destination apart from an unreadable one and reports the real inspection error
With the devui preview flag enabled (wiqd config flags set devui true), wiqd devui start now runs the server in its own visible console window on interactive, non-CI Windows runs unless --no-window is used; on macOS and Linux it still runs in the background. The new wiqd devui stop command shuts down servers started by this release cleanly. The server no longer holds a lock on its own install directory, so upgrades and uninstalls are not blocked. When upgrading from an earlier release, if a DevUI server is already running and the installer reports a locked file, end the node process listening on port 7317 or reboot, then retry. If its preferred trace port is busy, DevUI now listens on a free port and prints the exact workiq config set otlpEndpoint=... command needed to point tracing at it
Validate agents against declarative agent manifest v1.7 and v1.8, and offer both when completing the $schema field. Validation selects its rules from the manifest’s version field — set version to pin a schema version; $schema drives editor completion and hover only

Changes

Agent authoring guidance now covers the EmailActions and MeetingActions capabilities, which require declarative agent manifest v1.8 or later, and no longer pins a “latest” manifest version anywhere — it explains how to look up the current published version instead, so the guidance stays correct as new versions ship

0.11.0 — 2026-08-03

Features

Build a plugin end to end with the wiqd plugin commands, including OpenPlugin import and export
Sign in to Microsoft 365 directly through wiqd auth login, so authentication no longer needs a separate CLI installed

Fixes

Report upstream tool timeouts clearly, with recovery guidance that matches the actual failure
Accept an agent-relayed EULA consent without naming a version, and accept more natural URL phrasing
Recognize an agent project even when its marker file name differs in casing
Honor the EULA you accept during install, so the installer finishes setting up the Copilot plugin
Refresh the Copilot and Claude plugin during wiqd update, so skills no longer go stale after an upgrade
Fix 26 defects across the wiqd plugin lifecycle, so plugin create no longer overwrites an existing provisioned project
Route a standalone plugin request to the plugin commands instead of agent scaffolding, and restore the skill description so Copilot reliably picks up wiqd in a conversation

0.10.0 — 2026-07-30

Fixes

Fail fast with sign-in guidance before tenant-backed commands

0.9.0 — 2026-07-29

Features

Check for prerequisite CLIs and offer a consent-gated install
Notify you when a newer release is available, show what changed after an update, and browse version ranges with wiqd changelog --from/--to

Fixes

Seed extensions on first run, and repair them from the doctor, when a post-install step is blocked
Install the evaluation and Work IQ extensions as first-class dependencies
Scope the doctor’s antivirus exclusion advice to the project instead of a whole drive
Distinguish active from installed extensions in wiqd ext list, with clearer version reporting
Report upstream and precondition failures with actionable diagnostics
Describe the wiqd agent add command group properly in help output
apply —log-level to the shared logger before command actions

0.8.0-rc.1 — 2026-07-22

Features

Scaffold declarative agent projects from built-in templates
Add OpenAPI actions, skills, and authentication to an agent from the CLI
Validate agent manifests offline and check publish readiness before you ship
Provision an agent to a named environment and get its Copilot deep link
Package, share, inspect, publish, and delete agents across their full lifecycle
Initialize and run scored agent evaluations with concurrency and threshold controls
Ask, list, and monitor deployed agents from the terminal with structured output
Launch a local browser DevUI to build, debug, and evaluate agents on your machine
Drive the full build, improve, preview, and publish lifecycle from one guided Copilot skill
Create and edit agents safely with project checks and read-before-write safeguards
Generate and analyze evaluation suites with approval gates that protect the quality bar
Guide partners from a validated package through listing, certification, and go-live
Sign in, sign out, and inspect authentication across active extensions
Check for and install updates by version or channel, with dry-run support
Read installed release notes by version in text, JSON, or Markdown
Activate extensions and install Copilot, Claude, and editor integrations
Submit and browse product feedback from the CLI, with optional context
Emit versioned JSON success and error envelopes for scriptable use
Generate shell tab completion for Bash, Zsh, PowerShell, and Fish
Ship schemas and guides for extension manifests, OpenAPI, MCP, and OAuth