The Agent Skills Directory

[PROMPT_INJECTION]: The skill presents an indirect prompt injection attack surface because it processes potentially untrusted data from multiple sources to synthesize answers.
Ingestion points: Data enters the agent's context through tools like search-everything and ask-brain (which read from 'company memory'), as well as from the outputs of specialized agents like Mail/Gmail (which process external email content).
Boundary markers: Although the instructions tell the agent to "Synthesize the final answer with clear source boundaries," there are no specific directives instructing the model to ignore or neutralize malicious instructions embedded within the retrieved documents or emails.
Capability inventory: The skill utilizes call-agent to interact with other systems and can specifically query high-privilege information such as "secrets," "integration grants," and "workspace resources" through the Dispatch agent.
Sanitization: No input validation, escaping, or sanitization mechanisms are defined for the content ingested from external application agents or the Brain.
[NO_CODE]: This skill is comprised entirely of instructions in markdown format and does not include any executable scripts, binaries, or configuration files that could hide malicious logic.

ask-across-everything