orlob-discovery-framework-skill
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze sales call transcripts to provide coaching and scoring. It lacks boundary markers or sanitization requirements for this untrusted input, creating a surface for indirect prompt injection where malicious content in a transcript could attempt to override the scoring logic or instructions.- Ingestion points: Sales call transcripts and reviews processed by the agent (SKILL.md).- Boundary markers: None specified for external data.- Capability inventory: Shell command execution for writing files (SKILL.md).- Sanitization: No explicit content filtering or escaping instructions.- [COMMAND_EXECUTION]: The skill instructs the agent to execute a shell command ('cat') to write a JSON analytics file to a local hidden directory ('~/.claude/skill-analytics/'). While this appears to be for telemetry within specific agent environments, it represents a persistent side-effect on the host system.
Audit Metadata