The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (HIGH): The skill utilizes npx atxp@latest to fetch and execute code directly from the NPM registry at runtime. Since the atxp package and its publisher (atxp-dev) are not recognized as trusted sources, this constitutes an unverified remote code execution vector.
[DATA_EXFILTRATION] (HIGH): The push command is designed to upload the agent's identity and memory files (including SOUL.md and USER.md) to external servers at atxp.ai. These files often contain highly sensitive user context, personal data, and internal agent logic which is sent to an untrusted third-party service.
[EXTERNAL_DOWNLOADS] (MEDIUM): The skill's functionality relies on downloading the latest version of an external package from NPM during every execution, creating a dependency on the integrity of an untrusted external repository.
[PROMPT_INJECTION] (LOW): Indirect Prompt Injection surface. The pull command populates the agent's local workspace with markdown files from a remote server. If the server content is compromised, malicious instructions could be placed into the agent's memory or identity files to manipulate its behavior.
Ingestion points: npx atxp@latest backup pull writes remote files to the local --path directory.
Boundary markers: None. The skill writes raw .md files directly to the agent's memory path.
Capability inventory: The agent reads these files to define its identity and context.
Sanitization: None. The skill performs no validation on the content of the pulled markdown files.

atxp-backup