The Agent Skills Directory

[SAFE]: The skill contains purely educational content and code snippets for prompt engineering. It does not perform any automated actions or execute code on the host system outside of the provided documentation examples.
[INDIRECT_PROMPT_INJECTION]: The skill provides templates for processing potentially untrusted user data (e.g., receipt text, email threads, or text summaries). However, it actively promotes the use of security boundaries such as XML tags and triple backticks to delimit untrusted input from instructions, which is a recommended mitigation strategy against prompt injection attacks.
Ingestion points: User-provided strings in variables like receipt_text, text_to_summarize, and email_text (SKILL.md).
Boundary markers: The skill explicitly demonstrates using <receipt> XML tags and triple backticks (```) as delimiters.
Capability inventory: No system capabilities (filesystem, network, subprocess) are utilized; the snippets focus on string formatting for LLM prompts.
Sanitization: The skill includes guidelines for handling adversarial inputs and defining behavior for missing data fields.
[COMMAND_EXECUTION]: Code snippets provided are for prompt templating and basic logic; no commands are executed on the host system, and there are no shell interaction patterns.
[DATA_EXFILTRATION]: No network operations, hardcoded credentials, or access to sensitive local files (like SSH keys or AWS configs) were found in the skill content.

prompt-engineering