What is a plugin?

What is a plugin?

Caution

Alpha. The wiqd plugin command tree is enabled by default, but the surface is still alpha and the interface may change without notice.

wiqd plugin is the primary, superordinate developer construct in wiqd. A plugin is a Microsoft 365 / Teams app package — the same appPackage/ infrastructure wiqd already scaffolds, not a new artifact type — and it is the source-of-truth artifact: everything wiqd knows about a plugin’s capabilities is derived from it, not from any wiqd-side cache.

A declarative agent is one of the capabilities a plugin composes, not a separate parallel surface. A standalone wiqd agent project is simply the special case of a plugin whose single capability is a declarative agent — structurally the two are the same app package. Think in terms of the plugin first, and reach for the declarative agent as the capability that gives the plugin its conversational surface.

A plugin composes capabilities

A plugin can carry one or more of these, declared as top-level components in the app manifest:

Capability	Manifest location	Added by
An agent	copilotAgents.declarativeAgents (→ declarativeAgent.json)	wiqd plugin add agent
Skills	agentSkills[] (each a SKILL.md folder)	wiqd plugin add skill
Connectors	agentConnectors[] (remote MCP servers)	wiqd plugin add connector

A declarative agent brings its own API plugins (actions[]) and MCP references along with it — those live inside the agent’s own manifest, not as separate top-level plugin components.

A wiqd-owned descriptor, wiqd.plugin.json, marks the project root and caches the capability list for fast enumeration (used by wiqd plugin list). It is an advisory cache — the app manifest (appPackage/manifest.json) is always the source of truth, and commands like wiqd plugin show and wiqd plugin validate read the manifest itself rather than trusting the cache.

wiqd plugin create pins the scaffolded manifest to manifestVersion: "1.29" — the numbered (GA) Teams schema that supports agentSkills (1.28) and, at 1.29, url-only agentConnectors (where mcpToolDescription becomes optional) — so composed capabilities validate against a stable schema rather than devPreview.

The build loop

Capabilities compose into a single app package, in any combination, through one lifecycle — the same lifecycle a declarative-agent-only plugin uses:

wiqd plugin create --name my-plugin                 # scaffold from the ATK blank app template
cd my-plugin
wiqd plugin add agent                               # add a declarative agent component
wiqd plugin add skill --name "Triage Issues"        # add a SKILL.md skill
wiqd plugin add connector --name "Tools" \
  --description "Toolbelt over MCP" --url https://tools.contoso.com/mcp
wiqd plugin validate                                # static MVL check of the DA surface
wiqd plugin provision                               # register with Microsoft 365 (writes env/.env.<env>)
wiqd plugin package                                 # build a deployable .zip
wiqd plugin validate --mode deep                    # full app-package validation via ATK/AVL
wiqd plugin share --scope users --email you@contoso.com

Already have a plugin authored for another host? wiqd plugin import converts an Open Plugin (or Claude/Cursor plugin) into a wiqd plugin project instead of starting from create, and wiqd plugin export does the inverse. See the build-a-plugin walkthrough for the full step-by-step.

Three things wiqd calls a skill

The word “skill” is overloaded across wiqd — it means three genuinely different things depending on which command you’re looking at. Confusing them is the most common source of friction when you’re new to the plugin construct.

#	Surface	What it does	Lives in
1	wiqd plugin add skill	Scaffolds a SKILL.md folder and registers it in the plugin’s own app manifest (agentSkills[])	A standalone plugin project (wiqd.plugin.json)
2	wiqd agent add skill	Adds a skill to an existing declarative agent project’s own skill surface, via ATK (atk add skill)	An agent project (m365agents.yml)
3	Extension-contributed Copilot skills	The SKILL.md packages a wiqd extension bundles, that Copilot CLI itself loads as part of the wiqd@wiqd plugin	A wiqd extension’s skills/ directory, merged into the wiqd@wiqd plugin at install time

1. wiqd plugin add skill — a plugin capability

This scaffolds appPackage/skills/<slug>/SKILL.md and adds an entry to the manifest’s agentSkills[] — one of the three top-level capabilities a standalone plugin composes (see above). It requires a plugin project and does not require an agent component to be present; a plugin can be skill-only.

2. wiqd agent add skill — an agent’s own skill

This shells out to atk add skill to add a skill directly into an existing agent project, in-place. It produces the same kind of artifact (a SKILL.md folder referenced from agentSkills[]), but the command operates on an agent project rather than a standalone plugin, and it additionally supports --from <path> to import an existing skill directory or .zip, and --expose-to-copilot to expose the skill to mainline M365 Copilot. Use this when you’re extending an agent you already have; use wiqd plugin add skill when you’re composing a new, independent plugin from scratch.

3. Extension-contributed Copilot skills

This is a different axis entirely — it’s not about the M365 app package at all. A wiqd extension (the packages that contribute wiqd CLI commands, like the Agents Toolkit or Work IQ extensions) can ship its own SKILL.md files that teach GitHub Copilot CLI itself how to drive that extension’s commands conversationally. Those skills are merged into the single bundled wiqd@wiqd Copilot plugin at install time — they have nothing to do with agentSkills[] or any M365 app package.

Go deeper

• wiqd plugin — the full command reference
• Build a plugin — a step-by-step walkthrough
• wiqd plugin validate — static MVL vs. package-first deep validation
• Declarative agents — the agent capability a plugin can compose
• Host vs extensions — how wiqd itself is structured (a different use of “extension”)