A reusable AI prompt and structured review skill for identifying realistic failure scenarios before they cost you money, time, or trust.
Built around the philosophy behind The Cartographer Method: map before you build, because the failure modes are already there. You just haven't named them yet.
failure-map walks through any system - code, workflow, automation, or AI deployment - and produces a structured failure analysis written as if the incidents have already happened.
Not a bug hunt. Not a security audit. A pre-mortem exercise that surfaces what a future editor, operator, or API change could break without realizing it.
It works beyond code. That's the point.
| Domain | Examples |
|---|---|
| Code and APIs | ordering dependencies, shared state, version-coupled behavior |
| Business workflows | handoff gaps, approval bypass, scope drift |
| Automation systems | trigger-state coupling, silent failures, retry storms |
| AI deployments | hallucination surface, memory staleness, overreach risk |
| Data pipelines | trust drift, null propagation, transformation assumptions |
Most systems don't break because someone made an obvious mistake. They break because someone made a reasonable change without knowing what depended on the thing they touched.
failure-map forces that question before the change happens.
flowchart LR
A[System or workflow] --> B[Map fragility]
B --> C[Write plausible future incidents]
C --> D[Identify impact]
D --> E[Choose hardening action]
E --> F[Test]
E --> G[SOP]
E --> H[Alert]
E --> I[Refactor]
failure-map/
├── README.md
├── LICENSE.md
├── prompts/
│ └── failure-map.md <- the main prompt
└── examples/
├── content-publishing-workflow.md
├── ecommerce-erp-sync.md
└── ai-memory-layer.md
failure-map is a prompt-based skill. Run it with Claude, ChatGPT, or any capable LLM that supports long-context instruction following.
- Open
prompts/failure-map.md - Give the AI the system you want reviewed - paste code, describe a workflow, share a process doc, or describe an automation
- Let it map the failure surface
- Use the output to decide what becomes a test, an SOP, an alert, or a refactor
No special tooling required. The output is a markdown report you keep alongside the system it documents.
See the /examples folder for worked failure maps across three real-world scenarios:
- content-publishing-workflow.md - An AI-assisted content creation, review, scheduling, and distribution workflow
- ecommerce-erp-sync.md - An ecommerce-to-ERP order sync integration
- ai-memory-layer.md - A persistent memory layer for an AI agent system
Originally inspired by Matthew Honnibal's claude-skills pre-mortem command, then rewritten and expanded for operational workflow, AI, and systems-integration review.
MIT. Use it, fork it, adapt it.
Built by Todd Campbell
Allen T. Consulting
The Cartographer Method: Map. Stabilize. Deploy.
