The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to update project-level instruction files, specifically AGENTS.md and .github/instructions/conventions.instructions.md. These files directly influence the agent's behavior and constraints in future sessions.
Ingestion points: The skill ingests the entire current conversation history to identify significant learnings.
Boundary markers: It does not specify the use of delimiters or 'ignore' instructions when writing new content to these files, relying on standard markdown edits.
Capability inventory: The skill utilizes Read, Write, Edit, and Glob tools to modify the local project file system.
Sanitization: No explicit logic is provided to sanitize or filter the extracted content from the conversation history before it is persisted to the agent's instructions, creating a surface where session data can influence future system-level guidance.

wrap-up