The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/test_staircase.py uses the __import__('pathlib') function to dynamically load the pathlib module. While this occurs within a unit test for managing temporary directories, dynamic imports are often flagged as they can be used to hide the loading of unauthorized or malicious code.
[INDIRECT_PROMPT_INJECTION]: The skill ingests 'DPL events' from a DuckDB lake (synced from S3). These events contain external data such as URLs, metadata, and canonical links. The skill is instructed to output the resulting markdown report verbatim to the user. If the source data contains malicious instructions or formatting intended to manipulate the LLM, these would be included directly in the agent's output context. The skill lacks explicit boundary markers or sanitization specifically for the markdown output to prevent the agent from obeying instructions embedded in the report data.
[DATA_EXPOSURE]: The skill is designed to read from ${XDG_CONFIG_HOME:-~/.config}/agentic-analytics/bucket.json. This file contains sensitive environment configuration including S3 bucket names and AWS profiles. While this is the intended purpose of the skill to function, it represents access to sensitive local configuration files.
[COMMAND_EXECUTION]: The skill uses python3 -c shell snippets and local Python scripts to perform date arithmetic, database probing, and report generation. This involves executing several shell commands with variables derived from local configuration and environment variables.

staircase-report