The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell scripts (scripts/init-skeleton.sh and scripts/init-research-profile.sh) to generate the project skeleton and optional research profiles. These scripts use standard file system operations and do not perform network activities or access sensitive system paths.
[COMMAND_EXECUTION]: The skill configures a PostToolUse hook in .claude/settings.json that executes auto-format-python.sh. This hook uses ruff to automate code formatting and linting for Python files. The script safely parses environment-provided JSON input using an inline Python command.
[SAFE]: The skill implements a Phase 2 interactive workflow for filling out CLAUDE.md. While this involves reading project metadata (like package.json or README.md) which could contain untrusted data, the skill mitigates indirect prompt injection risks by requiring explicit user confirmation via AskUserQuestion before writing any generated content to disk.

init-project