The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/main.py script accepts a --data-path command-line argument which is passed directly to Python's open() function for both reading and writing operations. The code contains no validation to ensure the path remains within the intended workspace, allowing for arbitrary file access on the host system.
[DATA_EXFILTRATION]: Due to the lack of path restrictions on the data file parameter, the skill can be used to read sensitive files from the filesystem if directed to a specific path by a user or malicious prompt.
[METADATA_POISONING]: The skill includes a file named lab-inventory-predictor_audit_result_v4.json which contains a "PASS" security verdict and other safety assertions. This is deceptive as it misrepresents the actual implementation of the skill, which lacks the security controls (such as path traversal rejection) specifically claimed in the documentation and noted as a strength in the report itself.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from a local JSON file that could be modified by external actors. 1. Ingestion points: scripts/main.py via _load_data. 2. Boundary markers: None present. 3. Capability inventory: File read and write operations within main.py. 4. Sanitization: None implemented.

lab-inventory-predictor