The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The instructions encourage users to install and run a third-party package continues from NPM using npx continues or npm install -g continues. This package is not from a verified organization or well-known service.
[DATA_EXFILTRATION]: The tool's core functionality involves scanning and reading sensitive session data from multiple local directories, including ~/.claude/, ~/.cursor/, ~/.copilot/, and others across 16 different AI tools. These files typically contain conversation history, internal reasoning, and source code fragments.
[COMMAND_EXECUTION]: The skill facilitates the execution of various CLI tools (e.g., gemini, codex, cursor) and supports passing arbitrary flags or arguments to these tools via the --in and passthrough mechanisms.
[PROMPT_INJECTION]: The skill implements Indirect Prompt Injection by design. It extracts historical data from external files and injects it as a structured markdown document into a new AI agent session.
Ingestion points: The skill reads session data from local files such as .jsonl, .json, and SQLite databases (e.g., ~/.claude/projects/, ~/.cursor/projects/).
Boundary markers: Uses standard Markdown headers like ## Context and ## Recent Messages to separate injected content, but lacks instructions to the receiving agent to ignore embedded commands.
Capability inventory: The tool performs file reads across the home directory, writes markdown files, and executes shell commands to launch destination AI tools.
Sanitization: No sanitization of the transferred content is mentioned; the skill prioritizes preserving the full context of the original session.

continues-cli-session-transfer