The Agent Skills Directory

[SAFE]: The skill implements a robust lead scoring logic across multiple files, focusing on intent and deal size. No malicious patterns were detected.
[COMMAND_EXECUTION]: The skill uses a platform-specific tool, swan-update-skill, to modify its own Markdown instructions. This is used exclusively for persisting configuration settings and rules derived during the setup phase, rather than executing arbitrary shell commands.
[DATA_EXFILTRATION]: Data flows are strictly limited to the connected CRM system and designated alert channels for reporting purposes. No unauthorized data exfiltration to third-party domains was observed.
[PROMPT_INJECTION]: The skill is designed to ingest and process external data such as CRM notes and meeting transcripts. While this represents a surface for indirect prompt injection, the skill includes human-in-the-loop safeguards, such as human-readable CRM notes and audit trails.

score