The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is designed to read and export the complete local history of AI interactions (Claude Code, Codex, Kimi CLI). This history data may contain highly sensitive information, including source code, business logic, and potentially credentials or secrets previously shared in conversations.
[EXTERNAL_DOWNLOADS]: The setup command in scripts/history_index.py downloads native libsimple binary archives from GitHub (github.com/wangfenjin/simple). The skill implements security best practices by validating the downloaded files against hardcoded SHA-256 hashes before extraction.
[DYNAMIC_EXECUTION]: The skill uses SQLite's extension loading capability (sqlite3.load_extension) to load the native libsimple library at runtime. This is used to enable specialized Chinese language tokenization for history indexing and search.
[COMMAND_EXECUTION]: The skill executes various system commands using subprocess. This includes git for workspace status reporting, sysctl for monitoring system memory pressure during heavy embedding tasks, and native search tools like rg (ripgrep) or grep for efficient file scanning.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted historical conversation data. While the scripts are focused on data retrieval and recovery, the agent may be exposed to instructions embedded in past conversations. The documentation includes explicit guidance on auditing recovered content for security and privacy.

read-claude-code-history