The Agent Skills Directory

[PROMPT_INJECTION]: The skill acts as an orchestrator that passes artifacts (such as plans and implementation details) between different child agent instances. While this creates a potential surface for indirect prompt injection, the skill specifically mitigates this risk by requiring an 'Independent Review' phase. This phase uses a fresh model instance with no prior context to inspect the results against the original goal.
[COMMAND_EXECUTION]: The deployment component utilizes standard Git commands (git pull, git commit) to manage project state and verify revision history. These operations are scoped to the local repository's authorized upstream and are only executed after explicit authorization from the user for deployment tasks.
[DATA_EXFILTRATION]: The skill reads local configuration files and environment variables to identify available models and load user preferences. The documentation contains explicit instructions for users not to store credentials or API keys in these files, and the skill only interacts with model selectors natively exposed by the host coding environment.

relay