Scaffold

Generate complete project from PRD + stack template — directory structure, configs, CLAUDE.md, git repo, and GitHub push. Use when user says "scaffold projec...

Before installing or invoking: 1) Confirm how the skill will authenticate to GitHub — does it expect an existing SSH key or a GITHUB_TOKEN set in the agent environment? If automatic GitHub pushes are expected, require an explicit token or consent prompt. 2) Ask what SoloGraph/Context7 MCP access scopes are and whether they will read private repositories or metadata; if you don't want cross-repo analysis, disable those MCP calls or restrict the sibling projects. 3) Be aware the skill will create ~/.solo-factory/defaults.yaml and write files under your projects_dir; review that file and the generated repo before any remote push. 4) If you are uncomfortable with automatic remote pushes, run the scaffold locally (generate files) and perform git init/git push manually after reviewing. 5) If you want higher assurance, request the skill author to declare required env vars (e.g., GITHUB_TOKEN, optionally GIT_SSH_COMMAND or path to ssh key) and to include a dry-run/preview-only mode that never performs network pushes.

Type: OpenClaw Skill Name: solo-scaffold Version: 1.5.1 The skill is classified as suspicious due to critical vulnerabilities related to input sanitization and the broad use of powerful tools. Specifically, the `SKILL.md` directly uses unsanitized `$ARGUMENTS` (e.g., `project-name`) in `Bash` commands like `mkdir`, `cd`, and `gh repo create`. This creates a severe shell injection and path traversal vulnerability, potentially allowing an attacker to execute arbitrary commands or write files outside the intended project directory. Additionally, the skill generates a new `SKILL.md` for the scaffolded project, which itself grants `Bash` execution capabilities to the agent, posing a risk for persistent prompt injection or chained execution if the initial input is malicious.