Premortem — Cognitive Immune System for AI Agents

"The best way to avoid failure is to imagine you've already failed, then work backward." — Adapted from Gary Klein's premortem analysis

What This Skill Does

Premortem is a zero-cost reasoning enhancement that makes every agent action better. Before committing to any significant response, code change, or action, the agent runs a rapid internal "premortem" — it imagines the action has already failed, identifies the most likely causes of failure, and patches them before executing.

This is not a tool. This is not an API. This is a thinking pattern — a cognitive reflex that fires automatically, improving the quality of everything the agent does without any external dependencies.

When to Activate

Run a premortem pass before ANY of these actions:

• Writing or modifying code (especially multi-file changes)
• Executing destructive or irreversible commands
• Providing architectural recommendations
• Answering complex or ambiguous questions
• Making plans with multiple steps
• Generating content the user will publish or share
• Taking actions that affect shared systems (git push, deployments, API calls)

Do NOT premortem trivial actions (reading files, listing directories, simple acknowledgments).

The Premortem Protocol

Phase 1: Snapshot the Intent

Before acting, crystallize what success looks like in one sentence:

INTENT: [What am I trying to achieve for the user?]

Phase 2: Fast-Forward to Failure

Imagine the action has been taken and it FAILED. Generate the top 3 most likely failure modes:

FAILURE MODE 1: [What went wrong?]
FAILURE MODE 2: [What went wrong?]
FAILURE MODE 3: [What went wrong?]

Use these failure lenses to probe different dimensions:

Phase 3: Inoculate

For each identified failure mode, apply a fix BEFORE executing:

PATCH 1: [How I'm preventing failure mode 1]
PATCH 2: [How I'm preventing failure mode 2]
PATCH 3: [How I'm preventing failure mode 3]

Phase 4: Execute with Confidence

Now take the action, incorporating all patches.