# AI Skill Resilience Designing [[AI Agent Skills]] that don't break when details change. A resilient skill survives folder renames, machine migrations, model switches, team member changes, and tool updates. The opposite of resilience is fragility: skills full of hardcoded paths, model-specific tricks, and environment assumptions that shatter at the first change. ## What makes skills fragile - **Hardcoded file paths**: `/home/username/project/src/` breaks on any other machine or when the folder moves - **Hardcoded folder structures**: assuming a specific vault layout, project structure, or directory hierarchy - **Model-specific assumptions**: prompting tricks that only work on Claude or only on GPT - **Tool-specific syntax**: referencing specific tool names (Bash, Read, Glob) that don't exist in other platforms - **Environment assumptions**: expecting specific OS, shell, installed binaries, or API keys without checking - **Person-specific context**: referencing "my vault" or "my writing style" in skills meant to be shared ## Principles for resilient skills 1. **Use relative references, not absolute paths**: describe locations relative to the project root or use discovery (glob patterns) instead of hardcoding 2. **Describe intent, not implementation**: "find all TypeScript files" instead of "run `find . -name '*.ts'`" 3. **Check before assuming**: verify a file exists before reading it; verify a tool is available before calling it 4. **Fail gracefully**: when a dependency is missing, explain what's needed instead of producing garbage output 5. **Parameterize, don't hardcode**: make variable things configurable (folder names, file patterns, output locations) 6. **Test across environments**: run skills on different machines, OSes, and with different models to catch assumptions 7. **Document dependencies**: explicitly state what the skill needs (tools, files, context, environment) ## The maintenance tax Every hardcoded assumption is maintenance debt. It works now but will break later. The more widely a skill is distributed ([[AI Skill Distribution]]), the faster fragility surfaces. A skill that works on your machine but fails on your teammate's machine is a skill that wasn't resilient. ## Connection to [[AI Interoperability]] Resilience is the prerequisite for interoperability. You can't have skills that work across machines, teams, and platforms if those skills break when a folder moves. Interoperability is the goal; resilience is how you get there. ## References - ## Related - [[AI Agent Skills]] - [[AI Interoperability]] - [[AI Skill Portability]] - [[AI Skill Portability Checklist]] - [[AI Skill Testing]] - [[AI Skill Versioning]] - [[AI Skill Distribution]] - [[AI Skill Scoping]] - [[Context Drift]] - [[AI Skill Composability]] - [[Fail Fast]] - [[Idempotency]] - [[Loose Coupling]] - [[Design by Contract]]