The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (HIGH): The skill utilizes npx atxp@latest across all primary commands. This practice downloads and executes the most recent version of the 'atxp' package from npm at runtime without version pinning or integrity verification, allowing for arbitrary code execution if the package is compromised.
[DATA_EXFILTRATION] (MEDIUM): The push command facilitates the transfer of local .md files to external atxp.ai servers. Although restricted to markdown, these files frequently contain sensitive agent state, identity documents (e.g., SOUL.md), and private interaction logs.
[COMMAND_EXECUTION] (MEDIUM): The skill relies on shell command execution via the npx binary to perform its core functions.
[EXTERNAL_DOWNLOADS] (MEDIUM): The skill requires the installation of the @zvec/zvec package from an external registry for local indexing and search capabilities.
[PROMPT_INJECTION] (LOW): As an indirect prompt injection surface, the skill ingests untrusted data via memory pull (server files) and provides them to the agent via memory search. Ingestion points: File restoration from cloud and local search results. Boundary markers: None specified. Capability inventory: File writing (pull), network upload (push), and shell execution (npx). Sanitization: No evidence of sanitization for retrieved content.

atxp-memory