The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because its primary function is to ingest and analyze untrusted external artifacts such as sports reports, notebooks, and data tables.
Ingestion points: Figures, tables, notebooks, and reports provided by the user for audit (SKILL.md).
Boundary markers: Absent; there are no specific delimiters or instructions for the agent to ignore potentially malicious instructions embedded within the analytical artifacts.
Capability inventory: The skill includes a Python script (scripts/figure_audit_template.py) that can create directories and write files to the local file system.
Sanitization: Absent; the content extracted from the untrusted artifacts is used to populate audit checklists and could potentially be passed as arguments to the helper script.
[COMMAND_EXECUTION]: The skill includes a bundled Python script (scripts/figure_audit_template.py) which performs file system operations using pathlib.Path.write_text and mkdir. While the script is intended for generating audit templates, the destination path is determined by the --out command-line argument. This could be exploited to overwrite files or create directories in unauthorized locations if the agent is influenced by malicious input during the audit workflow.

anti-slop-analytics