skills/lovstudio/bp-skill/lov-bp/Gen Agent Trust Hub

lov-bp

Pass

Audited by Gen Agent Trust Hub on Oct 16, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, potentially untrusted project materials to generate investor narratives, which introduces a vulnerability to indirect prompt injection.
  • Ingestion points: The skills/bp-outline/SKILL.md file specifies that the skill ingests data from repositories, product documentation, PRDs, websites, analytics exports, and customer notes.
  • Capability inventory: The skill has extensive file-writing capabilities and executes Python scripts to process and audit the ingested data.
  • Boundary markers: The skill instructions attempt to structure output into specific Markdown files like brief.md and outline.md, but there are no explicit instructions for the agent to ignore potentially malicious commands embedded within the source materials.
  • Sanitization: No sanitization or escaping logic is implemented in the provided Python scripts to filter out instructions from the ingested text before it is processed by the agent.
  • [COMMAND_EXECUTION]: The skill executes local Python scripts included in the package to manage the user's workspace and perform structural audits.
  • Evidence: Multiple SKILL.md files (root and bp-polish) contain instructions to run python3 "$SKILL_DIR/scripts/init_bp.py" and python3 "$SKILL_DIR/scripts/audit_bp.py".
  • These scripts are used for deterministic tasks such as workspace initialization and outline linting, but they represent the execution of local code by the agent platform.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 16, 2026, 04:22 AM