The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a local Python Zip Application (briesearch.pyz) to perform operational tasks such as enforcing search budgets (budget-check), verifying report grounding (ground-check), and managing research file layouts (research-layout). These executions are scoped to research metadata management and do not involve unsanitized user input in a way that suggests command injection.
[DATA_EXFILTRATION]: While the skill interacts with external search providers (Tavily, Exa, etc.), it contains rigorous data protection guidelines in references/safety.md. It mandates the sanitization of URLs to remove credentials, the use of one-way digests for URL tracking, and strictly prohibits the inclusion of literal private code or user data in external queries.
[INDIRECT_PROMPT_INJECTION]: The skill explicitly acknowledges the risk of malicious instructions embedded in web content. In references/safety.md, it provides mandatory rules to ignore directives found within fetched content and to treat external text strictly as data for evidence, rather than as instructions for the agent's behavior.
[PROMPT_INJECTION]: Static analysis flagged potential instruction overrides in references/safety.md. Review of the content confirms these are defensive instructions designed to protect the agent from external attacks (e.g., advising the agent to ignore "Ignore previous instructions" strings found in web results), which is a security best practice.

briesearch