The Agent Skills Directory

[DYNAMIC_EXECUTION]: The slots.py script dynamically loads and executes Python code from a relative path using the importlib.util module.
Evidence: The _pipeline() function in slots.py constructs a path to ../pipeline/deals.py and uses spec.loader.exec_module(mod) to load and run its contents.
Risk: This pattern creates a dependency on code outside the skill's immediate directory. While intended for sharing logic between sibling skills (e.g., 'pipeline'), it allows for potential code execution if the external path is modified or points to an unintended file.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted text from contact replies to automate scheduling tasks, creating a surface for indirect injection.
Ingestion points: The parse command in slots.py accepts untrusted message text via the --text argument.
Boundary markers: The skill's instructions in SKILL.md explicitly mandate a user confirmation step ("Reply yes or no") before any calendar event is booked or invite is sent.
Capability inventory: The skill is capable of writing .ics files to the filesystem and modifying CRM data (deal stages) via the shared pipeline module.
Sanitization: The script uses regular expression matching (TIME_RE, DAY_RE, DATE_RE) to extract specific temporal patterns, which effectively filters out non-date instructions, although it does not sanitize the input for other downstream uses.

scheduler