The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a data processing workflow that ingests external CSV files. * Ingestion points: The script scripts/pretrade_liquidity_cost_modeling_diagnostics.py reads data from an input.csv file provided as a command-line argument. * Boundary markers: There are no explicit delimiters or instructions to ignore embedded content within the processed data. * Capability inventory: The skill is capable of reading and writing local files, but it does not have network access or the ability to execute arbitrary shell commands. * Sanitization: Although the script coerces data values to numeric and datetime formats, it explicitly includes the raw list of column names from the input file in its JSON summary output (summary["columns"]). This allows an attacker to inject text into the agent's context by naming CSV columns with malicious instructions.
[SAFE]: The skill's primary functionality is focused on legitimate financial modeling. Analysis of the scripts and instructions revealed no signs of credential exfiltration, persistence mechanisms, obfuscation, or unauthorized privilege escalation.

pretrade-liquidity-cost-modeling