The Agent Skills Directory

[DATA_EXFILTRATION]: Exposure of user's local directory structure. The SKILL.md file contains multiple absolute file paths pointing to a specific user's home directory (/Users/sargupta/SahayakAIV2/...), which reveals the local username and internal folder hierarchy.\n- [INDIRECT_PROMPT_INJECTION]: Vulnerability surface for instructions embedded in project management data. \n
Ingestion points: task.md and gemini.md (read via local Python scripts).\n
Boundary markers: None present; content from external files is parsed and presented directly to the agent.\n
Capability inventory: The skill can read file content and run local scripts (audit_task_health.py, generate_status_report.py) to process project data.\n
Sanitization: No sanitization or escaping is performed on the data extracted from the Markdown tasks before it is presented to the agent.\n- [COMMAND_EXECUTION]: Execution of local scripts with file path arguments. The skill relies on executing Python scripts provided in the package to parse Markdown files. These scripts take a file path as an argument, which could potentially be manipulated by an agent to access files outside the intended project scope, although the scripts themselves only perform regex parsing and printing.

scrum-master