chatgpt-journal-export
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from ChatGPT export files.
- Ingestion points: Processes
conversations.jsonor ZIP files provided by the user inscripts/extract.mjs. - Boundary markers: Extracted text is written to Markdown files without specific delimiters to mark the content as untrusted if the agent reads it later.
- Capability inventory: Performs local file writes via Node.js
fsmodule. - Sanitization: No sanitization is performed on the message content, as the intent is a verbatim export.
- [COMMAND_EXECUTION]: The skill instructs the agent to execute a bundled Node.js script locally.
- Evidence: The script
scripts/extract.mjsis executed using the localnoderuntime. The code analysis confirms it only uses built-in modules (fs,path,zlib) and performs no network operations. - [SAFE]: The skill includes several security-positive behaviors.
- Evidence: It explicitly uses restrictive file permissions (0o700 for directories, 0o600 for files) to ensure exported data is only readable by the current user. It also provides clear instructions to the agent to respect user privacy and avoid unauthorized data transmission.
Audit Metadata