The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is designed to query the agent's internal SQLite database at ~/.t3/userdata/statev2.sqlite. This file contains the complete history of command executions, thread titles, and messages, potentially exposing sensitive interaction data to the agent's current context.
[COMMAND_EXECUTION]: Several system commands are executed by the skill's scripts to manage the local environment:
git: Extensively used for worktree management, checking branch status, and comparing commits.
sqlite3 and python: Used to query the local state database.
lsof and du: Used to detect active processes in directories and calculate disk usage.
[DYNAMIC_EXECUTION]:
The documentation suggests writing and executing temporary Python scripts for complex database queries.
The skill utilizes the ELECTRON_RUN_AS_NODE environment variable to execute tool fallback commands via the Electron binary.
The maintenance scripts dynamically construct and invoke shell commands based on the contents and state of the local file system.
[INDIRECT_PROMPT_INJECTION]: The skill analyzes historical thread data, which represents a potential attack surface for indirect prompt injection.
Ingestion points: Historical messages and thread titles are read from the state database for activity scanning.
Boundary markers: No delimiters or safety instructions are used when processing this historical text.
Capability inventory: The skill possesses capabilities for file system modification (recursive deletion) and system command execution.
Sanitization: The scripts use regular expressions to identify specific patterns but do not sanitize the ingested text before processing.
[EXTERNAL_DOWNLOADS]: The worktree garbage collection script performs git fetch origin --prune to synchronize with remote repositories, which involves a network connection to the configured Git remote.

t3-code