The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The utility script scripts/post_explore_summary.py references a non-standard PyPI registry (pypi.yonyon.ai) to obtain the yg-mcp-core package. This introduces potential supply chain risk through the use of an unverified third-party repository.
[INDIRECT_PROMPT_INJECTION]: The skill ingests codebase files and search results using Read, Grep, and Glob, passing this data to several sub-agents for analysis. The instructions do not define boundary markers or provide sanitization, creating a surface where malicious instructions in the code being analyzed could influence agent behavior.
Ingestion points: src/**, package.json, and search outputs defined in SKILL.md.
Capability inventory: Bash, Write, and Agent tool execution for sub-agent creation.
Boundary markers: None present in the prompts for exploration sub-agents.
Sanitization: No evidence of filtering or escaping external content before interpolation into prompts.
[METADATA_POISONING]: The skill contains deceptive metadata and prose, claiming compatibility with non-existent future model versions (e.g., "Opus 5.5") and specifying a future release date of "April 2026". Such claims can mislead users about the skill's tested environment and current capabilities.
[COMMAND_EXECUTION]: The Bash tool is used to execute several local components, including a repository structure indexer hook and Python/Node.js utility scripts within the skill directory.
[DYNAMIC_EXECUTION]: The skill makes extensive use of the Agent tool to concurrently spawn multiple specialized sub-agents with prompts that are dynamically constructed based on user input and file system content.

explore