The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/build_hints.py allows the execution of arbitrary local commands via the --llm-ideas-cmd argument. The command string is split using shlex and executed using subprocess.run to integrate with external LLM CLI tools.
[DYNAMIC_EXECUTION]: The skill uses subprocess.run to dynamically execute a command string provided at runtime. While it uses shlex.split to mitigate shell injection, the feature allows for the execution of any binary available on the host system.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external files (JSON and CSV) such as market summaries and news headlines. This data is incorporated into a JSON payload and piped to an external command (typically an LLM CLI). Malicious content within the source data could influence the output of the downstream tool.
Ingestion points: The script reads data via read_market_summary, read_anomalies, and read_news_reactions in scripts/build_hints.py.
Boundary markers: The payload sent to the external command lacks explicit delimiters or instructions to ignore embedded commands in the data fields.
Capability inventory: The script has the ability to execute shell commands (subprocess.run) and write to the filesystem (Path.write_text).
Sanitization: Textual content from external sources, such as news headlines, is not sanitized before being passed to the external command.

edge-hint-extractor