The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides a framework for an agent to process conversation turns and local markdown files to extract and categorize durable collaboration context. This creates an attack surface where malicious instructions embedded in the conversation history or in the memory files could influence the agent's extraction and classification behavior.
Ingestion points: The scripts/memory_manifest.py script reads .md files from a user-specified directory, and the references/prompt-template.md file defines a prompt that ingests recent conversation slices via the <recent_messages> placeholder.
Boundary markers: The prompt template uses tag-like placeholders (e.g., <recent_messages>, <memory_manifest>) but lacks explicit instructions to treat the ingested content strictly as data or to ignore any embedded commands.
Capability inventory: The skill includes a Python script capable of scanning directories and reading file contents, combined with a prompt that directs the agent to modify its long-term state based on the ingested content.
Sanitization: No sanitization, validation, or escaping of the ingested data is performed before it is presented to the language model for processing.

memory-extractor