The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructs the agent to consult external framework documentation at bootstrap time to ensure it uses the most current CLI flags and templates. This creates a vulnerability surface for indirect prompt injection where malicious content in external docs could attempt to subvert agent behavior.
Ingestion points: External framework documentation accessed during Phase 1 (SKILL.md).
Boundary markers: Absent; no specific instructions are provided to the agent to ignore or delimit instructions found within the external documentation.
Capability inventory: Full shell access for project configuration and deployment, including pnpm, git, wrangler, and gh (SKILL.md, references/).
Sanitization: Absent; no explicit filtering or validation of external documentation content is defined.
[REMOTE_CODE_EXECUTION]: The skill executes various official scaffolding tools (e.g., pnpm create cloudflare@latest, uv init, cargo new) and recommends the use of remote GitHub Actions (jdx/mise-action). While these are from well-known and reputable sources, they involve downloading and executing code from the internet during the project initialization process.
[COMMAND_EXECUTION]: The skill involves extensive use of shell commands to manage files, initialize git repositories, pin runtimes with mise, and wire security hooks through hk. It specifically checks for successful command execution by attempting "deliberately bad commits" to verify that security gates are functional.
[EXTERNAL_DOWNLOADS]: The skill fetches templates and dependencies from official registries including npm, PyPI, and crates.io. It also references external tools for deployment and CI configuration.

bootstrap-project