The Agent Skills Directory

[Command Execution] (MEDIUM): Path Traversal vulnerability in initialization logic.
Evidence: In scripts/init_skill.py, the variable skill_path is constructed using os.path.join(squads_dir, name), where name is a raw command-line argument. A user could provide a relative path like ../../some-sensitive-dir to create directories and files outside of the intended squads/ directory.
[Data Exposure] (LOW): Hardcoded developer environment paths.
Evidence: Scripts scripts/add_config_awareness.py and scripts/add_mcp_awareness.py contain hardcoded absolute paths: /Users/yuranikolaev/Developer/antigravity/ag-skill-factory/squads. This leaks the developer's local username and file structure.
[Indirect Prompt Injection] (LOW): Susceptibility to injection via unvalidated template variables.
Ingestion points: scripts/init_skill.py accepts <skill-name> and populates SKILL.md templates.
Boundary markers: None.
Capability inventory: File writing, directory creation, and shell script generation.
Sanitization: No validation or escaping is performed on the name argument before it is used in path construction or string replacement within the skill template.

skill-creator