The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill retrieves model configuration and session guidance from official Claude documentation domains (claude.com and code.claude.com). This allows the agent to provide accurate recommendations for model selection and effort levels based on current service capabilities.
[COMMAND_EXECUTION]: The skill uses dynamic context injection to execute read-only git commands, including git branch, git log, and git status, during its initialization. This information is used to ground the AI's understanding of the repository's current state and recent activity.
[COMMAND_EXECUTION]: Shell utilities such as mktemp are utilized to generate temporary HTML files for decision tables. These files are stored in system-scoped temporary directories and are used solely as a scannable rendering surface for complex rounds of questions.
[SAFE]: The skill implements strong security and logic controls, such as an 'auto-guard' to prevent the AI from making critical design decisions without user confirmation and a 'stop-on-gap' rule to prevent the synthesis of incomplete information.

interview