The Agent Skills Directory

[PROMPT_INJECTION]: The skill exposes an attack surface for indirect prompt injection through its custom factor expression field.
Ingestion points: User-supplied input via the expr field in the SKILL.md qsh-form is interpolated into the agent's prompt template.
Boundary markers: The prompt template lacks explicit delimiters or instructions to the agent to disregard instructions potentially embedded within the user's expression.
Capability inventory: The associated script scripts/decay_analysis.py has the capability to read local parquet files, write JSON reports, and interact with the pandadata market data API.
Sanitization: There is no evidence of validation or sanitization for the content of the expr field.
[COMMAND_EXECUTION]: The skill provides scripts/decay_analysis.py, which is intended for command-line execution. It performs file system operations including reading from data/factors and writing results to data/decay_reports.
[DYNAMIC_EXECUTION]: The script scripts/decay_analysis.py dynamically modifies sys.path at runtime to import the pandadata_runtime module from a relative path (../skill-pandadata-api/scripts). This creates a hard dependency on the local environment's folder structure and the presence of a specific sibling skill.

factor-decay