The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructions include a check for a local file at ~/.claude/company-context.md. This is used to gather existing context about the company to provide tailored advice. This access is local to the agent environment and does not involve network transmission.
[COMMAND_EXECUTION]: The skill utilizes local Python scripts provided within the skill package (scripts/feature_prioritizer.py, scripts/product_health_scorer.py, scripts/product_portfolio_analyzer.py) to analyze product data. These scripts perform mathematical calculations on JSON input and do not execute external or arbitrary commands.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect injection as it processes user-provided JSON files. * Ingestion points: Data is ingested through the --input flag of the provided Python scripts which read user-supplied JSON files. * Boundary markers: The scripts use the standard json library for parsing, which provides structural isolation of data fields. * Capability inventory: The Python scripts are restricted to data processing and logging; they do not have capabilities for network access, file system modification, or process spawning. * Sanitization: Input is validated through JSON parsing and numeric type casting during the scoring process.

cpo-advisor