The Agent Skills Directory

[SAFE]: The skill operates as a document review tool, analyzing spec-product.md files to identify missing product definitions.
[COMMAND_EXECUTION]: The skill uses local shell commands (ls, grep, node, mkdir) to resolve file paths and read configuration from stelow.json. These operations are limited to the current project directory and are used for legitimate workflow management.
[REMOTE_CODE_EXECUTION]: While the skill mentions launching 'parallel reviewers' or 'subagents', it refers to internal AI orchestration patterns defined in the references/ directory. No external script downloads or piped bash executions were detected.
[DATA_EXPOSURE]: The skill reads project-specific configuration (stelow.json) and product specifications. It does not attempt to access sensitive system files (like .ssh or .aws) or exfiltrate data to external domains.
[PROMPT_INJECTION]: The instructions contain 'Golden rules' and mode definitions for the AI agent, but these are legitimate behavioral guidelines for the critique process and do not attempt to bypass platform safety filters.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided spec-product.md files. While this is an ingestion surface, the skill implements a 'Validate-Then-Repair' pattern and structured checklists that reduce the risk of following instructions embedded in the reviewed data. (Severity: LOW)

stelow-product-plan-critique