The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill contains a hardcoded cryptographic secret (dd05f1c54d63749eda95f9fa6d49v442a) used to generate API signatures. This secret is present in multiple utility scripts (references/scripts/datasource_ops.py, references/scripts/proc_ops.py, references/scripts/linkage_ops.py) and documented in references/signing-datasource-guide.md.
[COMMAND_EXECUTION]: The script references/scripts/gen_all_comps.py utilizes subprocess.run() with shell=True to pipe the dashboard preview URL to the Windows system clipboard using the clip command. This represents a risk of command injection if the input strings are manipulated.
[DYNAMIC_EXECUTION]: The references/scripts/proc_ops.py utility uses subprocess.run() to load and execute other Python scripts within the skill folder (comp_ops.py) using arguments derived from user input or configuration files.
[EXTERNAL_DOWNLOADS]: The skill instructions (references/dataset-guide.md) and scripts (references/scripts/proc_ops.py) direct the user to install the pymysql package via pip to enable database connectivity. Installing unverified packages at runtime introduces supply chain risks.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a broad attack surface for indirect prompt injection as it is designed to ingest untrusted data from diverse sources—including SQL databases, external REST APIs, and user-uploaded files—and render this content within the agent's dashboard context. There is no evidence of robust sanitization or boundary marking for this ingested content across the provided scripts.

jimubi-dashboard