The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill gathers free-form text from the user during the project initialization process to populate .planning/PROJECT.md. Since this document is intended to serve as a primary context source for other AI agent skills, it presents an attack surface for indirect prompt injection if malicious instructions are included in the project description.
Ingestion points: SKILL.md Phase 3 (Deep Questioning) collects user descriptions.
Boundary markers: The collected input is stored directly in markdown files without specific delimiters or isolation instructions.
Capability inventory: The skill suite has broad capabilities including file system access and command execution via bash and the GitHub CLI.
Sanitization: No specific sanitization or filtering is applied to the user's project description before it is stored in the project documentation.
[COMMAND_EXECUTION]: The skill automates project initialization using a series of bash commands and local helper scripts to manage git repositories, create directory structures, and setup complex worktree layouts.
Evidence: SKILL.md executes git init, mkdir, and gh repo create. It also invokes scripts/setup-worktrees.sh which performs significant file system restructuring, including moving the git directory to a bare repository and removing files from the project root to create a worktree structure.
[EXTERNAL_DOWNLOADS]: The skill creates a GitHub Actions workflow that references established external actions from trusted organizations.
Evidence: .github/workflows/release.yml references actions/checkout@v4, actions/setup-node@v4, and softprops/action-gh-release@v2. These are standard, well-known components used for CI/CD automation.

kata-new-project