The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill follows a pattern where it parses values from external files (such as Fidelity CSVs and Buy Tickets) and then modifies the source code of a local Python script (fin-guru-private/strategies/dividend_margin_monte_carlo.py) by interpolating these values into the file. This script is then executed using uv run python. This "script generation and execution" workflow is inherently risky as it can be exploited for code injection.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from untrusted or external sources which creates an attack surface for indirect prompt injection.
Ingestion points: The skill reads "Buy Ticket" markdown files from fin-guru-private/fin-guru/tickets/ and Fidelity positions CSVs from notebooks/updates/.
Capability inventory: The skill has the capability to read files, write/modify local scripts, and execute shell commands via uv run.
Boundary markers: The parsing logic uses ticker whitelists and attempts to cast values to floats, which provides some protection.
Sanitization: The skill lacks robust sanitization for inputs that are eventually interpolated into executable code, creating a path from data ingestion to code execution if the agent does not strictly validate the parsed data before script modification.
[COMMAND_EXECUTION]: The skill uses uv run python to execute scripts that have been modified at runtime. While the execution targets local files within the project structure, the dynamic nature of the scripts being run increases the potential impact of a successful injection attack.

monte-carlo