The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes a workflow where an agent reads terminal output and performs actions based on it, creating a vulnerability to instructions embedded in that output.
Ingestion points: Terminal data is ingested via phux snapshot --json @N as described in SKILL.md.
Boundary markers: There are no instructions to use delimiters or ignore commands found within terminal snapshots.
Capability inventory: The agent possesses write and execution capabilities through phux run, phux send-keys, and phux paste.
Sanitization: The skill does not specify any sanitization or validation of the terminal content before it is processed by the agent.
[COMMAND_EXECUTION]: The skill is designed to execute arbitrary commands in persistent terminal environments.
The phux run command is used for discrete execution in target terminal panes.
Interactive inputs and multiline text are managed through the send-keys and paste verbs.
[DATA_EXFILTRATION]: Terminal snapshots provide access to potentially sensitive session history.
The agent uses phux snapshot to capture the state of terminal panes, which may contain environment variables, command history, or sensitive data displayed during the session.

using-phux