How to use the Pre-Mortem Plan Challenge agent skill

What is an agent skill, and what does the challenge skill do?

An agent skill is a folder containing a SKILL.md file that gives an AI agent instructions for a specific task. When a skill is installed, the agent can load those instructions when a request matches the skill's description. The challenge skill is a set of instructions for analysing a plan.

The skill uses the pre-mortem technique. You imagine the plan has failed spectacularly a year from now, then work backwards to explain why. The source frames this as a way to make a plan survive contact with reality, not a way to kill it.

• Extracts the core assumptions the plan relies on, such as market, customer, execution, competitive, financial and dependency assumptions.
• Rates each assumption for confidence (High, Medium, Low or Unknown) and for impact if wrong (Critical, High, Medium or Low).
• Flags vulnerabilities: low-confidence assumptions with high impact.
• Maps the dependency chain, showing which assumptions rely on others and the weakest link.
• Assesses reversibility, separating reversible bets from irreversible commitments.
• Sets kill switches: what would have to be true at 30, 60 and 90 days to continue versus kill or pivot.
• Lists hardening actions: validation steps, alternative approaches and contingencies.

How do you install the challenge skill?

You can install the skill in two ways. Both start by downloading the skill's ZIP file from the page where you found this guide.

For Claude Code, unzip the file into your personal skills folder at ~/.claude/skills/, or into a project's .claude/skills/ folder if you want the skill available only in that project. For the Claude apps, upload the ZIP under Settings → Capabilities → Skills.

• Download the challenge skill ZIP from this page.
• Unzip it so that the challenge folder, containing SKILL.md, sits inside ~/.claude/skills/ or a project's .claude/skills/ folder.
• Or, in the Claude apps, open Settings → Capabilities → Skills and upload the ZIP.

How do you trigger the challenge skill?

The skill's description lists the situations where it applies: before significant resource commitment, before presenting to a board or investors, when feedback has been one-sidedly positive, or when there is pressure to move fast and figure it out later. You can ask in plain language that describes one of these situations, and paste in the plan.

The source also defines a slash command, /em:challenge <plan>. Use it if your setup exposes that command. Otherwise, a natural request works.

• Pre-commitment: "Before we spend the Q3 budget on this launch plan, run a pre-mortem on it: [paste plan]"
• Board prep: "I'm presenting this expansion plan to the board Thursday. Challenge it and find the weak assumptions: [paste plan]"
• Positive feedback: "Everyone on the team loves this hiring plan. Play devil's advocate and stress-test it: [paste plan]"
• Pressure to move fast: "We want to start building this roadmap next week. Pre-mortem it first: [paste plan]"

What output should you expect?

The skill asks the agent to produce a Challenge Report named after your plan. It is a structured vulnerability map, not a verdict. The source says its output is not permission to stop; it lets you make conscious decisions about which risks to validate, hedge or accept.

The report follows a fixed layout, so you can scan it quickly and compare it across plans.

• CORE ASSUMPTIONS (extracted), each with confidence and impact ratings
• VULNERABILITY MAP, with critical risks (act before proceeding) and high risks (validate before scaling)
• DEPENDENCY CHAIN, with the weakest link identified
• REVERSIBILITY ASSESSMENT, listing reversible bets and irreversible commitments
• KILL SWITCHES, with continue and kill/pivot conditions at 30, 60 and 90 days
• HARDENING ACTIONS, with specific validation steps, alternatives and contingencies

How do you get the most useful results?

The quality of the report depends on the detail in your plan. A one-line summary produces generic assumptions. Include the goal, the timeline, the budget or headcount involved, the customers you are targeting, and the external parties you depend on.

The skill includes plan-type patterns for product roadmaps, go-to-market plans, hiring plans and fundraising plans. Naming your plan type helps the agent apply the right questions. The source also lists hard questions to ask, such as "What's the bear case, not the base case?" and "What would an enemy of this plan attack first?"

• Name the plan type (roadmap, go-to-market, hiring, fundraising) so the matching questions are applied.
• Give concrete numbers where you have them, such as runway, deal cycle length or headcount plan.
• Ask the agent to challenge the plan again after you have validated the top risks.
• Treat ratings as the agent's judgement, and check market and financial facts yourself.

What does the skill need, and what about safety?

The skill lists no requirements. It does not run scripts and does not use the network. The package contains a single file, SKILL.md.

Your plan text is sent to the agent for analysis. If the plan contains confidential data you would not share with an AI service, do not paste it in. Remove or generalise sensitive details first.

The skill is not a substitute for expert legal, financial or regulatory advice. It surfaces questions; the agent can be wrong about market or financial facts.

• No listed requirements; no scripts are run; no network use.
• Plan text is sent to the agent, so redact confidential figures and names first.
• Use it alongside qualified advisers for legal, financial and regulatory questions.