deepworkplan

SUSPICIOUS. The core planning/onboarding behavior is coherent and mostly same-org, but the skill’s footprint extends into transitive installation of other skills and optional third-party agent tools, which is broader than a simple router and raises trust-chain risk. I do not see confirmed credential theft or malware behavior in the provided files, but the installation/delegation surface makes this a medium-to-high security risk skill.

No clear malware, data exfiltration, or covert persistence is evident. The main security concern is export path traversal: an absolute evidence_path in ledger data can cause export to copy a local file and write it outside the requested destination. Gate command execution is an intended feature but must be used only with trusted commands and contracts. The fragment as shown also appears syntactically invalid due to empty exception class bodies.