The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/audit_partners.py executes shell commands via subprocess.run to call curl. The target URLs for these probes are extracted from the local repository's PartnersGrid.tsx file, creating a surface where malicious configuration data could influence command execution.
[DYNAMIC_EXECUTION]: The audit_partners.py script employs a dynamic module loading pattern using __import__('os') to access environment variables. Although used for legitimate environment configuration, dynamic imports are a common vector for obfuscating malicious activity.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it reads and processes data from external files (partner metadata and URLs) to perform network requests and file updates without explicit boundary markers or sanitization.
Ingestion points: The skill reads partner definitions from app/(main)/(home)/PartnersGrid.tsx and localized taglines from JSON files in src/i18n/messages/.
Boundary markers: None identified; the skill parses TSX and JSON files using regular expressions and standard JSON loaders without using delimiters or instructions to ignore embedded commands.
Capability inventory: The skill has the ability to write to files (add_partner.py, normalize_logo.py), execute network probes via curl (audit_partners.py), and process images using the Pillow library.
Sanitization: Data parsed from the repository is interpolated directly into files and command arguments without escaping or sanitization logic.

lov-maintain-partners