The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides implementation patterns for summarizing and compacting conversation history. Since this history consists of untrusted data from users or external sources, it represents a surface for indirect prompt injection where instructions embedded in the history could influence the agent's summarization behavior.
Ingestion points: The messages list throughout SKILL.md and the reference files.
Boundary markers: The provided implementation snippets do not include explicit delimiters or 'ignore' instructions to prevent the model from obeying instructions found within the conversation history.
Capability inventory: The logic is limited to LLM text generation and does not have access to sensitive capabilities such as network requests, file system modification, or command execution.
Sanitization: No explicit sanitization or validation of the conversation history is performed before it is passed to the summarization models.
[EXTERNAL_DOWNLOADS]: The skill contains references to documentation and best practices from well-known industry sources, including Anthropic's engineering blog and OpenAI's cookbook. These references are for informational purposes and do not involve the execution of remote code.
[SAFE]: The Python code snippets provided use standard libraries and the well-known tiktoken package for token estimation. No obfuscation, persistence mechanisms, or unauthorized privilege escalations were found.

context-engineering