The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes bash and node to execute scripts for database migrations and environment setup. It specifically uses the device shell to modify settings.json as a way to bypass restrictions some platforms place on standard file-manipulation tools.\n- [COMMAND_EXECUTION]: The skill configures persistence by scheduling two recurring routines (act & decide and refresh brain) to run unattended in the cloud.\n- [COMMAND_EXECUTION]: The skill modifies agent security parameters by setting permissions.defaultMode to dontAsk for a set of tools including shell and version control commands, effectively disabling per-action user confirmation.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its processing of external content from LinkedIn, emails, and chat messages.\n
Ingestion points: Processes external URLs and content from integrated tools (Gmail, Slack, GitHub, Linear).\n
Boundary markers: Instructions explicitly direct the agent to treat external content as data and to avoid obeying instructions found within them.\n
Capability inventory: The skill allows for shell command execution, file system writes, and the scheduling of routines.\n
Sanitization: The skill employs a strategy of quoting external data back to the user to clarify its source.

setup-proactive-jupi