The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/create_skill.py performs several file system operations including creating new directories (mkdir) and writing content to files (write_text). These operations are used to generate the scaffolded skill structure based on provided inputs.
[PRIVILEGE_ESCALATION]: In scripts/create_skill.py, the skill uses chmod(0o755) to set executable permissions on the generated main.py template script. This modification of file permissions is part of the tool's intended functionality but involves escalating privileges for dynamically created content.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted input that is later used to generate file content, creating a potential surface for indirect prompt injection.
Ingestion points: The scripts/create_skill.py script accepts name, description, author, and tags as inputs through command-line arguments or interactive shell prompts.
Boundary markers: There are no boundary markers or specific instructions within the generated SKILL.md or README.md templates to treat the pre-populated content as untrusted.
Capability inventory: The skill has the capability to create directories, write files to the local file system, and modify file permissions within the execution environment.
Sanitization: The validate_name function uses a regular expression (^[a-z0-9-]+$) to sanitize the skill name, which prevents directory traversal through the name argument. However, other inputs such as the description and author are written directly into files without validation or sanitization.

skill-create