Conversation Export

Pass

Audited by Gen Agent Trust Hub on Mar 10, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses shell commands, specifically bash and jq, to read and extract data from Claude Code's local JSONL session files. It also utilizes rm for individual file deletion during batch management tasks.
  • [EXTERNAL_DOWNLOADS]: The integration guide recommends the installation of external Model Context Protocol (MCP) servers from well-known organizations, such as the official Notion MCP server.
  • [DATA_EXFILTRATION]: The skill is designed to transmit conversation content to external platforms (Notion, Feishu). This operation is conducted through the user's own pre-configured MCP infrastructure.
  • [PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection because it ingests and processes untrusted conversation history for summarization and export.
  • Ingestion points: Reads active conversation memory and session JSONL files (SKILL.md).
  • Boundary markers: Employs Markdown headers and separators to distinguish between user and assistant content (references/format-templates.md).
  • Capability inventory: Possesses file write access and network communication via MCP tools (SKILL.md).
  • Sanitization: Documentation specifies HTML entity escaping for code content in HTML exports (references/format-templates.md).
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 10, 2026, 04:57 AM