The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process the agent's conversation history from src/memory/working-memory.ts and CLAUDE.md to perform summarization and fact extraction. This creates a surface where malicious instructions embedded in previous conversation history could attempt to influence the compaction process or future agent behavior when the summary is re-injected as context.
Ingestion points: The agent reads conversation history from src/memory/working-memory.ts and episodic memory from CLAUDE.md (as specified in the 'Summarize & Compact Flow' and 'Full Reset Flow').
Boundary markers: The instructions do not specify the use of delimiters or 'ignore embedded instructions' warnings when processing the history through the summarization prompts.
Capability inventory: The skill utilizes read_file and write_file (SKILL.md) to modify memory files, episodic memory (CLAUDE.md), and potentially interact with a local SQLite database.
Sanitization: There is no mention of filtering or sanitizing the ingested conversation history before it is passed to the summarization and extraction prompts (prompts/memory/summarizer.toon, prompts/memory/extractor.toon).

add-clear