The Agent Skills Directory

[SAFE]: The skill follows security best practices for data handling and integrity.\n
Implements SHA-256 integrity verification in scripts/build_landscape.py to validate search artifacts.\n
Uses atomic file operations (temporary files and os.replace) to ensure file system consistency.\n
Explicitly prohibits gathering sensitive information such as credentials, API keys, or private environment data.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes research papers from external sources, which constitutes an attack surface for indirect prompt injection.\n
Ingestion points: Ingests paper titles, abstracts, and metadata from candidates_deduped.jsonl within research search bundles.\n
Boundary markers: Employs structured JSONL formats and strict schema validation for all incoming records.\n
Capability inventory: Capabilities are limited to local file system organization within docs/papers/ and .bensz-api/, with no use of dangerous functions like eval() or exec().\n
Sanitization: Performs title normalization (ASCII conversion and regex filtering) in scripts/catalog.py to sanitize potential injection payloads in identifiers.\n- [COMMAND_EXECUTION]: The skill includes an instruction to use gh api for reporting design defects.\n
This behavior is scoped to local logging by default and requires explicit user consent for external transmission, following legitimate developer bug-reporting workflows.

research-literature-radar