The Agent Skills Directory

[DATA_EXPOSURE]: The skill performs no network operations. Data processing is restricted to local file reads and writes within user-specified paths. The 'autoskill' migration logic lacks explicit path validation for file names in the source index, which is a common pattern for local tools but represents a potential path traversal vulnerability surface if used with malicious source directories.
[COMMAND_EXECUTION]: The script does not use subprocesses, shell commands, or dynamic execution functions like eval() or exec().
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external systems (ChatGPT and Claude Code memory files) that could contain malicious embedded instructions.
Ingestion points: scripts/import.py reads data through the parse_chatgpt, parse_claude_local, and _merge_autoskill_half functions.
Boundary markers: Imported content is structurally separated using Markdown headers and unique hidden markers (e.g., ).
Capability inventory: The skill possesses no capabilities for network access or arbitrary command execution.
Sanitization: Content is trimmed of whitespace, though no further sanitization of the processed text is performed.
[REMOTE_CODE_EXECUTION]: No remote code execution patterns, runtime code generation, or dynamic loading of external code were detected.
[EXTERNAL_DOWNLOADS]: The skill has no external dependencies and does not download any packages or scripts at runtime.

memory-import