The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it is designed to ingest and process untrusted user-supplied time series data.
Ingestion points: Data enters the agent's context through functions like analyze_time_series and create_ts_features defined in SKILL.md and referenced in the workflow.
Boundary markers: The instructions lack delimiters or explicit warnings to ignore any natural language instructions that might be embedded within the data values or timestamps.
Capability inventory: The skill utilizes powerful tools including arbitrary Python execution and extensive libraries (e.g., statsmodels, lightgbm, torch).
Sanitization: There is no evidence of data sanitization or input validation to filter malicious instructions before the data is processed or described by the agent.
[COMMAND_EXECUTION]: The skill provides and encourages the use of numerous Python code snippets for data analysis and model training.
Usage: Workflow steps in SKILL.md and implementations in reference files (e.g., forecasting-methods.md, deep-learning-ts.md) provide templates for executing code via the agent's tools to perform ML tasks.
[EXTERNAL_DOWNLOADS]: The skill references and depends on multiple third-party libraries.
Dependencies: Standard ML and data processing packages are utilized, including pandas, numpy, statsmodels, pmdarima, prophet, lightgbm, scikit-learn, torch, pytorch-forecasting, pytorch-lightning, darts, ruptures, tsfresh, and feast. These are established and well-known services.
[SAFE]: Eight of the reference files (e.g., references/architecture-patterns.md, references/security-best-practices.md) are largely comprised of 150 repetitive sections of generic technical text. While this content is mostly irrelevant to the skill's primary focus on time series (containing out-of-scope Rust and networking examples) and creates unnecessary context overhead, no specific malicious payloads were identified within these sections.

ml-time-series