The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection vulnerability surface. The skill instructs the agent to ingest and analyze output from terminal sessions (e.g., read screen, read scrollback). A malicious process, build log, or test output could contain hidden instructions designed to hijack the agent's logic or influence subsequent actions, such as deployment decisions.
Ingestion points: Terminal screen buffers and scrollback history are read into the agent's context during the 'Fan-Out' and 'Watcher' patterns.
Boundary markers: No specific boundary markers or instructions to ignore embedded commands within the terminal output are provided.
Capability inventory: The skill enables full command execution across multiple sessions, environment variable manipulation, and filesystem access via command injection into shells.
Sanitization: There is no evidence of sanitization or filtering of the content read from the sessions before it is processed by the agent.
[COMMAND_EXECUTION]: Extensive shell command execution capabilities. The skill is designed to manage parallel terminal sessions and execute arbitrary commands provided by the user or determined by the agent's orchestration logic. While this is the intended functionality, it provides a high-privilege surface for an attacker to exploit if they can influence the agent via the indirect prompt injection surface mentioned above.

multi-session