The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted user input from the PROJECT_OUTLINE argument to drive project configuration and script generation.
Ingestion points: PROJECT_OUTLINE argument in SKILL.md used to determine components and project context.
Boundary markers: None present. The agent is instructed to parse and follow the context provided in the outline without delimiters or safety instructions regarding embedded commands.
Capability inventory: The skill uses Bash to execute pnpm commands and Write/Edit to create scripts like dev.sh and package.json content.
Sanitization: No sanitization or validation is performed on the user-provided outline before it influences logic that generates executable content.
[COMMAND_EXECUTION]: The skill performs automated command execution on scripts it generates dynamically.
Evidence: The instructions in SKILL.md (Step 11) require the agent to run pnpm install and pnpm verify immediately after creating the project structure.
Risk: If a malicious PROJECT_OUTLINE tricks the agent into adding dangerous commands to the package.json scripts (e.g., inside the verify, test, or lint commands), those commands will be executed in the local environment during the bootstrap process.

init