The Agent Skills Directory

[SAFE]: The skill operates locally within the user's home directory (~/.ultrabrain/) and does not perform any unauthorized network communication or external data exfiltration. All data management is scoped to the vault and its associated configuration files.
[SAFE]: Configuration changes, such as the installation of 'SessionStart' and 'PreCompact' hooks, follow a strict 'Read → diff → confirm → Edit' discipline. This ensures that the user is fully aware of and must explicitly approve any modifications to ~/.claude/settings.json.
[SAFE]: The shell scripts provided for the hooks (session-start.sh and pre-compact.sh) are transparent and perform benign local tasks, such as counting files and reading logs to provide status updates to the agent. They do not contain any suspicious commands or hidden payloads.
[SAFE]: The skill's design includes multiple guardrails, such as collision handling for new captures and human-in-the-loop checkpoints during the ingestion and linting processes, preventing accidental data loss or unauthorized modifications.
[SAFE]: No obfuscation techniques, credential harvesting, or indirect prompt injection vulnerabilities that could bypass safety guidelines were found in the skill's instructions or scripts.

ultrabrain