microsoft-agent-365

Name: microsoft-agent-365
Author: vinayaklatthe

161stars34forks

How to install microsoft-agent-365?

vinayaklatthe/microsoft-security-skills/microsoft-agent-365

$npx -y skills add vinayaklatthe/microsoft-security-skills --skill microsoft-agent-365

Installs into the current project.

Summary

Guidance for managing AI agents at enterprise scale with Microsoft Agent 365 - the control plane that lets admins observe, govern, and secure every agent (Microsoft, Copilot Studio, and third-party) from a single registry, using Microsoft Entra Agent ID for identity, Microsoft Pu

Files · 1

View on GitHub

SKILL.md

1	# Managing AI Agents with Microsoft Agent 365
2
3	Microsoft Agent 365 is the control plane for agents - it extends the infrastructure you
4	use to manage people (identity, access, data protection, threat defense) to AI agents,
5	regardless of where they originate (Microsoft, Copilot Studio, partner/third-party including
6	agents built on Anthropic Claude, OpenAI, or Google Gemini, or custom-built). It organises the
7	work into three pillars: observe, govern, and secure.
8	Generally available for Commercial since 1 May 2026, licensed per user, with Microsoft E5 as
9	the recommended prerequisite.
10
11	## When to use
12	Standing up enterprise-wide management for a growing fleet of AI agents - when you need a single
13	inventory, consistent lifecycle and access controls, and unified security across agents built
14	on different platforms. Use it as the orchestration layer that ties together Entra, Purview,
15	and Defender for agents.
16
17	Do not use this skill for Purview data-security controls on a single agent in isolation (use
18	`purview-agent-365-security`), or to remediate Microsoft 365 Copilot oversharing of end-user
19	content (use `purview-copilot-oversharing` / `m365-oversharing`).
20
21	## Pick the right pillar for the problem
22	\| Goal \| Pillar \| Primary surface \|
23	\|---\|---\|---\|
24	\| Know how many agents exist, who owns them, how they're used \| Observe \| Agent registry + Agent Map in Microsoft 365 admin center \|
25	\| Onboard, approve, review, and retire agents with policy \| Govern \| Agent management in M365 admin center + Microsoft Purview \|
26	\| Give each agent a managed identity and risk-based access \| Secure \| Microsoft Entra Agent ID + Conditional Access \|
27	\| Stop agents leaking or over-accessing sensitive data \| Secure \| Microsoft Purview (DSPM for AI, DLP, sensitivity labels, IRM) \|
28	\| Detect malicious or unsafe agent behaviour at runtime \| Secure \| Microsoft Defender for Cloud Apps (real-time agent protection) \|
29
30	Rule of thumb: treat every agent as a first-class, governed identity - not a feature of an
31	app. The more autonomous the agent, the more it must be managed like a privileged account.
32
33	## Approach
34	1. Observe - build the inventory - Use the Agent 365 agent registry in the Microsoft 365
35	admin center to get a single, centralised view of every agent (Microsoft, Copilot Studio,
36	third-party), its owner, adoption, activity, and health. Use Agent Map to see
37	relationships and access.
38	Verify: the registry lists agents from more than one build platform with named owners.
39	2. Assign identity (Entra Agent ID) - Ensure each agent has a managed **Microsoft Entra
40	Agent ID** so it can be authenticated, authorised, and governed like any other directory
41	identity - not a shared secret or anonymous app.
42	Verify: each production agent resolves to a distinct Entra Agent ID with an owner.
43	3. Govern lifecycle & access - Onboard agents intentionally with IT oversight, apply
44	least-privilege access to resources/data, require approvals, and run periodic access reviews;
45	retire low-value or ownerless agents on a cadence.
46	Verify: an agent cannot reach a data source it was not explicitly granted.
47	4. Secure the data (Purview) - Bring agent instances into Microsoft Purview: DSPM for AI for
48	visibility, data classification + sensitivity labels, DLP, Insider Risk Management, and
49	Communication Compliance. New Agent 365 agent instances are auto-enabled for audit and
50	sensitive-data detection.
51	Verify: DSPM for AI's AI observability page shows the agent instance and its data risks.
52	5. Apply risk-based access (Conditional Access) - Use Entra Conditional Access so agents and
53	the users they act for meet consistent, risk-based access conditions.
54	Verify: a risky context blocks or steps up the agent's access as policy dictates.
55	6. Defend at runtime (Defender) - Use Microsoft Defender for Cloud Apps real-time agent
56	protection to detect and block unsafe behaviour, prompt-injection, and malicious activity
57	while the agent runs.
58	*Verify: a simulated unsafe action raises a Defender alert

People who liked this also viewed

6 picks

Type

microsoft-agent-365

161stars34forks

View on GitHub

How to install microsoft-agent-365?

vinayaklatthe/microsoft-security-skills/microsoft-agent-365

$npx -y skills add vinayaklatthe/microsoft-security-skills --skill microsoft-agent-365

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Managing AI Agents with Microsoft Agent 365
2
3	Microsoft Agent 365 is the control plane for agents - it extends the infrastructure you
4	use to manage people (identity, access, data protection, threat defense) to AI agents,
5	regardless of where they originate (Microsoft, Copilot Studio, partner/third-party including
6	agents built on Anthropic Claude, OpenAI, or Google Gemini, or custom-built). It organises the
7	work into three pillars: observe, govern, and secure.
8	Generally available for Commercial since 1 May 2026, licensed per user, with Microsoft E5 as
9	the recommended prerequisite.
10
11	## When to use
12	Standing up enterprise-wide management for a growing fleet of AI agents - when you need a single
13	inventory, consistent lifecycle and access controls, and unified security across agents built
14	on different platforms. Use it as the orchestration layer that ties together Entra, Purview,
15	and Defender for agents.
16
17	Do not use this skill for Purview data-security controls on a single agent in isolation (use
18	`purview-agent-365-security`), or to remediate Microsoft 365 Copilot oversharing of end-user
19	content (use `purview-copilot-oversharing` / `m365-oversharing`).
20
21	## Pick the right pillar for the problem
22	\| Goal \| Pillar \| Primary surface \|
23	\|---\|---\|---\|
24	\| Know how many agents exist, who owns them, how they're used \| Observe \| Agent registry + Agent Map in Microsoft 365 admin center \|
25	\| Onboard, approve, review, and retire agents with policy \| Govern \| Agent management in M365 admin center + Microsoft Purview \|
26	\| Give each agent a managed identity and risk-based access \| Secure \| Microsoft Entra Agent ID + Conditional Access \|
27	\| Stop agents leaking or over-accessing sensitive data \| Secure \| Microsoft Purview (DSPM for AI, DLP, sensitivity labels, IRM) \|
28	\| Detect malicious or unsafe agent behaviour at runtime \| Secure \| Microsoft Defender for Cloud Apps (real-time agent protection) \|
29
30	Rule of thumb: treat every agent as a first-class, governed identity - not a feature of an
31	app. The more autonomous the agent, the more it must be managed like a privileged account.
32
33	## Approach
34	1. Observe - build the inventory - Use the Agent 365 agent registry in the Microsoft 365
35	admin center to get a single, centralised view of every agent (Microsoft, Copilot Studio,
36	third-party), its owner, adoption, activity, and health. Use Agent Map to see
37	relationships and access.
38	Verify: the registry lists agents from more than one build platform with named owners.
39	2. Assign identity (Entra Agent ID) - Ensure each agent has a managed **Microsoft Entra
40	Agent ID** so it can be authenticated, authorised, and governed like any other directory
41	identity - not a shared secret or anonymous app.
42	Verify: each production agent resolves to a distinct Entra Agent ID with an owner.
43	3. Govern lifecycle & access - Onboard agents intentionally with IT oversight, apply
44	least-privilege access to resources/data, require approvals, and run periodic access reviews;
45	retire low-value or ownerless agents on a cadence.
46	Verify: an agent cannot reach a data source it was not explicitly granted.
47	4. Secure the data (Purview) - Bring agent instances into Microsoft Purview: DSPM for AI for
48	visibility, data classification + sensitivity labels, DLP, Insider Risk Management, and
49	Communication Compliance. New Agent 365 agent instances are auto-enabled for audit and
50	sensitive-data detection.
51	Verify: DSPM for AI's AI observability page shows the agent instance and its data risks.
52	5. Apply risk-based access (Conditional Access) - Use Entra Conditional Access so agents and
53	the users they act for meet consistent, risk-based access conditions.
54	Verify: a risky context blocks or steps up the agent's access as policy dictates.
55	6. Defend at runtime (Defender) - Use Microsoft Defender for Cloud Apps real-time agent
56	protection to detect and block unsafe behaviour, prompt-injection, and malicious activity
57	while the agent runs.
58	*Verify: a simulated unsafe action raises a Defender alert

People who liked this also viewed

6 picks

Type