The Agent Skills Directory

[METADATA_POISONING]: The skill exhibits deceptive metadata by claiming a 'read' tier in its YAML frontmatter despite performing multiple mutating operations like creating and deleting groups. Additionally, the author field identifies as 'longbridge', which is inconsistent with the author context of 'nickbuddy2046', potentially misleading users about the skill's official status.
[COMMAND_EXECUTION]: The skill is designed to execute shell commands via the longbridge CLI tool to interact with stock data and user account settings.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided strings (symbols, group names, prices) and interpolates them into shell commands, creating a potential injection surface.
Ingestion points: Stock symbols, price targets, and watchlist group names accepted across alert.md, sharelist.md, and watchlist.md.
Boundary markers: A mandatory two-step protocol (Preview then Wait) is required for every mutation, ensuring explicit user confirmation before any command execution.
Capability inventory: Subprocess calls to longbridge alert, longbridge sharelist, and longbridge watchlist subcommands.
Sanitization: Instructions suggest wrapping names in double quotes, but do not specify comprehensive filtering for shell metacharacters or delimiters.
[EXTERNAL_DOWNLOADS]: The skill documentation refers users to the official GitHub repository for longbridge-terminal to fulfill its requirement for the external CLI dependency.

stock-watchlist