throng

SUSPICIOUS. The stated purpose matches the capabilities: it is a delegation/orchestration skill for nested agent runs. The main concern is trust and data-flow opacity: the skill forwards repository context and prompts to external harnesses/models and may let them edit the live working tree, while the exact throng publisher, install source, and backend endpoints are not verifiable from the provided material. No direct credential theft, malicious pre-execution, or explicit exfiltration endpoint is shown, but the cross-agent data sharing and unverifiable backend provenance make this medium risk.