The Agent Skills Directory

[SAFE]: Extensive analysis across all files confirmed no malicious patterns or security vulnerabilities.
[DATA_EXFILTRATION]: The skill manages interactions with external APIs (e.g., Tibber, OpenWeather) but implements strict security controls.
The 'Atlas' and 'Ada' agents are explicitly instructed to store all credentials in 'secrets.yaml' rather than hardcoding them in configurations or code.
No patterns of unauthorized data collection or exfiltration to unknown domains were found.
[PROMPT_INJECTION]: The skill uses a multi-agent framework to organize complex tasks, which actually increases behavioral consistency.
Agent 'Souls' define specific communication personas that do not attempt to override system safety guidelines or platform constraints.
'Iron Laws' defined in the orchestrator provide technical guardrails for hardware and automation generation.
[COMMAND_EXECUTION]: The skill allows the use of shell tools for infrastructure management, but with high-integrity oversight.
The 'Forge' agent uses Bash for Docker management and system backups, with a strict 'no deploy without backup' policy.
No instances of arbitrary or suspicious command execution were identified.
[EXTERNAL_DOWNLOADS]: External references are limited to official documentation, well-known smart home platforms (Nabu Casa), and the author's own public repository.

aurora