The Agent Skills Directory

[SAFE]: The skill is a standard project generator that copies files from a template directory and performs placeholder replacement. No malicious behavior or suspicious code patterns were detected.
[COMMAND_EXECUTION]: The skill executes benign shell commands for project initialization and validation. Specifically, it runs bun install to set up dependencies and bun run check to perform type-checking and build verification using established tools like tsdown and publint.
[PROMPT_INJECTION]: The skill contains an indirect prompt injection surface through the ingestion of user-provided project metadata (e.g., package name, author). However, this is the intended functionality for project customization. The skill maintains high safety by explicitly instructing the agent not to automate publishing, login, or CI workflow creation, thereby limiting the impact of any potential injection.
Ingestion points: Project metadata fields (NAME, PROJECT_NAME, DESCRIPTION, AUTHOR, NPM_OWNER, REPO, HOMEPAGE, YEAR) in SKILL.md.
Boundary markers: Absent; placeholders are directly interpolated into the template files.
Capability inventory: Shell execution for bun, rg, and git commands documented in the scaffolding procedure.
Sanitization: Absent; the skill relies on the agent's standard interpolation behavior.
[CREDENTIALS_UNSAFE]: The template includes a NOTES.md file that describes how to use npm login and manage tokens. This is handled safely as the instructions are purely educational for the user and the skill body explicitly forbids the agent from executing these sensitive commands or automating the login process.

start-package