The Agent Skills Directory

[COMMAND_EXECUTION]: The workspace initialization script scripts/init-workspace.py uses subprocess.run() to execute a local script at ~/.codex/skills/agent-folder-init/scripts/scaffold.py. This is a design choice to delegate agent-specific folder setup to a related skill.
[EXTERNAL_DOWNLOADS]: The skill's primary function in SKILL.md is to run npx @shipshitdev/v0. This downloads and executes the author's scaffolding package from the public NPM registry. The package belongs to the verified skill author ('shipshitdev').
[INDIRECT_PROMPT_INJECTION]: The skill provides an attack surface for indirect prompt injection by ingesting and processing untrusted user data (product briefs and entity descriptions) to drive code generation, as described in references/legacy-manual-route.md.
Ingestion points: User-provided product descriptions, entity names, and field definitions.
Boundary markers: None explicitly defined in the manual intake instructions.
Capability inventory: The skill possesses file system write access and local command execution capabilities through its Python scripts.
Sanitization: There is no explicit sanitization logic shown for entity names or field descriptions before they are interpolated into code templates.
[SAFE]: All external service references in references/deployment-guide.md and elsewhere target well-known, trusted providers including Clerk (Auth), MongoDB Atlas (Database), and Vercel, Railway, and Render (Deployment). Hardcoded environment variable keys in the documentation are placeholders and do not contain actual secrets.

fullstack-workspace-init