The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the internet via the WebSearch tool without implementing specific boundary markers or sanitization, creating a surface for indirect prompt injection. 1. Ingestion points: The references/intelligence-gathering.md file specifies various WebSearch queries to collect company and stakeholder data. 2. Capability inventory: The skill is configured with Bash, Write, and WebSearch tools (referenced in SKILL.md), providing a path for potentially malicious instructions found online to interact with the system. 3. Boundary markers: There are no instructions defining delimiters or explicit guidelines to ignore instructions embedded in the search results. 4. Sanitization: The skill does not include steps to filter or validate retrieved web content before it is processed by the agent.

deal-closer-playbook