The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill architecture establishes a multi-agent pipeline that ingests untrusted user data (tasks, research queries) and processes it through agents with file system and shell execution capabilities. This structure creates an inherent surface for indirect prompt injection, where an attacker could embed malicious instructions in codebase files or research data that are then processed by the Planner or Coder agents. The skill specifically includes an "Adversarial Reviewer" role and structured YAML-delimited handoffs to mitigate these risks.
[DYNAMIC_EXECUTION]: The provided agents/setup.sh script is a utility for dynamically generating agent configuration files. It processes local markdown files and prepends tool-specific frontmatter to create environment-specific agent definitions. While the source templates are local, this pattern of automated configuration generation is a form of dynamic execution.

agent-collaboration