export-chat-history
Warn
Audited by Gen Agent Trust Hub on Jun 21, 2026
Risk Level: MEDIUMDATA_EXFILTRATIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill reads sensitive local data from directories used by Claude Code and Cursor to store chat history and project context.
- Evidence: Scans
~/.claude/projectsand~/.cursor/projectsto extract session transcripts and metadata. - [DATA_EXFILTRATION]: Aggregated session data is uploaded to a remote service via the GitHub CLI.
- Evidence: Executes the
gh gist createcommand to publish processed session files. - [COMMAND_EXECUTION]: Shell commands are constructed using variables derived from local directory names and file paths.
- Evidence: Interpolates
{project-dir}and$SESSION_FILEwithin commands likejq,gawk, andstat. - [PROMPT_INJECTION]: The skill ingests untrusted chat logs to generate titles and descriptions, creating an indirect prompt injection surface.
- Ingestion points: Local session logs in
~/.claude/projects/and~/.cursor/projects/. - Boundary markers: Absent.
- Capability inventory: File system access, shell execution, and network operations (GitHub CLI).
- Sanitization: Absent.
Audit Metadata