How to use the Adversary Engagement with MITRE Engage agent skill

What is an agent skill and what does this one do?

An agent skill is a folder containing a SKILL.md file with instructions, plus optional templates, reference material and scripts that an AI agent can load when a task matches the skill's description. This skill is for planning, not deployment. It teaches the agent to define a strategic goal, set measurable objectives, map the attacker techniques your team expects to see to matching decoy activities, and write rules of engagement.

The skill follows the MITRE Engage 10-Step Operational Process. It uses the three engagement goals (Expose, Affect, Elicit) and the strategy bookends Prepare and Understand. Instead of placing decoys at random, the agent works from a threat model toward specific activities.

• Defines a strategic goal in plain language and documents where deception fits your existing cyber strategy.
• Writes falsifiable, time-bound Operational Objectives for each engagement goal.
• Maps ATT&CK technique IDs to Engage Activity names and IDs, with the live matrix as the authoritative source.
• Drafts gating criteria and rules of engagement, including who can escalate and when to hand off to incident response.
• Runs a bundled Python script (scripts/process.py) that turns a list of ATT&CK technique IDs into a coverage report and a plan skeleton.

Install the skill from the ZIP

Download the ZIP for this skill from this page and unzip it. The skill folder should be named designing-adversary-engagement-with-mitre-engage and contain SKILL.md, assets/template.md, references/standards.md and scripts/process.py.

For a personal install, place the folder in your user skills directory, ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ directory so that it is available only inside that project.

If you use the Claude apps instead of a local setup, you can upload the skill under Settings → Capabilities → Skills.

• Personal: ~/.claude/skills/designing-adversary-engagement-with-mitre-engage/
• Project only: .claude/skills/designing-adversary-engagement-with-mitre-engage/ inside your project folder
• Claude apps: Settings → Capabilities → Skills, then upload the skill

Check the requirements before you start

The skill needs Python 3 to run scripts/process.py. It needs access to the live MITRE Engage matrix at engage.mitre.org for canonical Activity names and IDs. The bundled map is only a starter list, so do not treat it as the final reference.

The skill also expects a documented set of critical assets, a threat model or prioritized list of threat actors, and an existing detection and response capability. It assumes executive sponsorship and written legal review before any live engagement.

• Python 3 (required to run scripts/process.py)
• Access to https://engage.mitre.org/matrix/ for current Activity names and IDs
• A list of critical assets and plausible adversaries
• Working detection and response that engagement supplements
• Legal sign-off before running any engagement with live adversaries

How to trigger the skill with natural requests

You do not need special syntax. Describe the planning problem in plain language and the agent will load the skill when your request matches its scope. The skill is triggered by requests such as deciding whether to engage attackers, building a deception program, writing an engagement operation plan, or choosing deception activities against a specific threat actor.

Give the agent your strategic goal, your critical assets, and the ATT&CK techniques you expect, since those inputs drive the mapping.

• "We have honeytokens and canary tokens but no strategy. Help us write an adversary engagement plan for our source-code repo."
• "Leadership asks whether we should engage attackers. Walk through the legal, operational and resourcing questions."
• "Our threat actor uses T1078, T1021 and T1552. Which MITRE Engage activities should we consider?"
• "Draft Operational Objectives for Expose and Affect for our VPN, with falsifiable time-bound targets."

Example: map techniques to a coverage report

The bundled script takes ATT&CK technique IDs and produces a coverage report and plan skeleton. The source does not document the script's command-line flags, so check the script's own header or usage text before running it. The skill itself maps common techniques as follows.

The starter mappings in the skill are T1078 to Decoy Credentials and Lures, T1083 to Decoy Content and Pocket Litter, T1046 to Network Diversity and Decoy Systems, T1021 to Decoy Systems and Network Manipulation, and T1552 to Decoy Credentials and Artifact Diversity.

• Start from the techniques your threat model says the adversary uses.
• Confirm each Activity name and ID against the live matrix before writing it into the plan.
• Use assets/template.md as the structure for the final operation plan.

Safety notes: legal, scripts and network

The skill itself states that engagement operations with live adversaries need legal sign-off. Running them without it creates entrapment, evidence-handling and liability risk. The skill also limits Affect activities to your own network and says never to act on infrastructure you do not own.

The skill runs scripts (scripts/process.py) and is marked as not using the network. Review the script before running it on your machine, and keep in mind that the skill's own guidance says the live matrix must be consulted separately at engage.mitre.org.

• Get written legal review before any deployment.
• Affect activities apply only to your defender network.
• Do not treat the bundled Activity map as current; verify against the live matrix.
• Review scripts/process.py before executing it.

Tips for better results

Write objectives as measurable statements with a time limit, such as an alert target within a set number of minutes. A goal without an objective cannot be scored after the operation.

Treat Expose alerts as high-fidelity. A touch on a decoy usually indicates unauthorized activity, so connect those alerts to a response path before deploying anything.

• Name a specific asset and a specific adversary rather than asking for a generic program.
• Ask the agent to score the operation against your objectives in the after-action step.
• Use the deployment skills for honeypots, honeytokens, canary tokens and decoy files to implement the activities this skill selects.