semantica
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFE
Full Analysis
- [DATA_EXPOSURE]: The included preflight utility (
scripts/preflight.py) incorporates a robust protection mechanism that blacklists sensitive file paths and stems associated with credentials, such as.ssh,.aws,.env, andid_rsa. Additionally, the script uses secure POSIX file descriptor operations withO_NOFOLLOWto prevent symlink-based attacks during file inspection. - [REMOTE_CODE_EXECUTION]: The skill architecture avoids automated execution of upstream code. It uses a checksum-based verification process to ensure local file integrity against a pinned manifest (
references/source.json) before any operations are proposed to the user. - [INDIRECT_PROMPT_INJECTION]: While the skill facilitates the ingestion of external documents and data for knowledge graph construction—creating a potential surface for indirect prompt injection—it mitigates this through strict instructional constraints. It mandates explicit user approval for data flows and provides clear boundaries between offline code inspection and runtime service interaction.
- [SAFE]: No malicious patterns, obfuscation, or unauthorized persistence mechanisms were identified. The skill follows security best practices by limiting its own capabilities to read-only integrity checks and requiring user consent for sensitive actions.
Audit Metadata