The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): The skill instructions (Step 2) provide a shell command template that interpolates user-provided input (task_description) directly into a python execution string: python scripts/estimate_task.py --task "<task_description>". This creates a high risk of command injection, where a malicious user could provide a task description containing shell metacharacters (e.g., ;, &&, or backticks) to execute unauthorized commands.
[REMOTE_CODE_EXECUTION] (MEDIUM): The skill depends on a local script scripts/estimate_task.py which is not provided within the analyzed files. This is an unverifiable dependency that performs codebase analysis. Without the script source, it is impossible to verify if the script performs safe operations or contains malicious logic like data exfiltration.
[PROMPT_INJECTION] (LOW): This skill exhibits an indirect prompt injection surface (Category 8).
Ingestion points: User-provided task descriptions are captured and processed.
Boundary markers: The instructions use double quotes as delimiters, but do not provide logic for escaping or sanitizing internal quotes or shell metacharacters.
Capability inventory: The skill uses the bash tool with python, ls, find, and wc permissions.
Sanitization: There is no instruction to sanitize or validate the task_description before it is passed to the shell.

eta