The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes several shell-based tools including sqlite3, jq, find, sed, and sort to interact with the local filesystem and query application databases. These tools are used for the intended purpose of retrieving chat history.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process historical conversation data from state.vscdb (SQLite) and JSONL transcripts. Because these files contain past user-generated content and assistant outputs, they represent an attack surface where previously stored malicious instructions could influence the agent's current behavior when the history is read.
Ingestion points: Reads data from ~/Library/Application Support/Cursor/User/globalStorage/state.vscdb, ~/.cursor/projects/*/agent-transcripts/*.jsonl, and conversation-search.db.
Boundary markers: The skill uses markdown headers like ## USER and ### ASSISTANT to structure the output but does not include explicit instructions for the agent to ignore potential commands embedded within the historical text.
Capability inventory: The skill has the ability to execute shell commands via Bash for database querying and file manipulation (sqlite3, jq, find, sed). It does not include network writing or arbitrary file writing capabilities.
Sanitization: The instructions provide guidance on escaping JSON paths for sqlite3 (e.g., \$.field) and suggest using jq -r for text extraction to handle raw data safely.

exploring-cursor-history