The Agent Skills Directory

[SAFE]: No malicious patterns or behaviors were detected in the skill instructions or associated scripts. The skill aligns with its stated purpose of performing research and report generation.
[INDIRECT_PROMPT_INJECTION]: As a tool that processes external web data, the skill is inherently exposed to indirect prompt injection. Mitigation: Ingestion occurs through host-provided search and web reading tools (SKILL.md); boundary markers include mandatory quotation marks and bracketed source references (e.g., [S001]); capabilities are restricted to writing to a local delivery directory and utilizing host tools; sanitization is performed via requirements for exact quoting of source material and the preservation of source snapshots for verification.
[COMMAND_EXECUTION]: The skill includes a Python script (scripts/validate_research_bundle.py) for local validation of research results. A technical audit confirms the script uses only Python's standard library and performs benign structural and integrity checks. It implements robust security measures, including path sanitization to prevent directory traversal when verifying file snapshots. The subprocess usage identified by automated tools is located in the test suite and is not part of the runtime skill logic.
[DATA_EXFILTRATION]: The skill explicitly defines and enforces data privacy boundaries, forbidding the access or exfiltration of sensitive information such as credentials, tokens, or environment variables. It restricts file operations to a specific, user-designated delivery directory.

double6-deep-research