The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill identifies that data from sources like feedback logs and issue reports is untrusted. It includes specific defensive instructions directing the agent to ignore any embedded commands within that data, effectively mitigating the risk of indirect prompt injection.
[COMMAND_EXECUTION]: The skill invokes a local script (scripts/p2p_filesystem.py) to manage file paths and save reports. These operations are scoped to the project's local workspace and do not involve unauthorized privilege escalation or remote code execution.

retrospect