The Agent Skills Directory

[PROMPT_INJECTION]: The skill ingests untrusted content from project configuration files such as CLAUDE.md without explicit boundary markers or sanitization protocols. This creates an indirect prompt injection surface where instructions embedded in the analyzed data could influence agent behavior, particularly when combined with the skill's file-writing and command execution capabilities.
[COMMAND_EXECUTION]: The instructions utilize standard shell commands including ls, pwd, and mkdir to audit project directory structures and create new skill directories as part of its optimization workflow.
[SAFE]: The provided project templates for API servers, web apps, and enterprise environments use standard placeholders for configuration variables (e.g., sk_test_..., DATABASE_URL) and do not contain hardcoded sensitive credentials or malicious external references.

claude-guide