The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a persistent memory system that stores arbitrary user-provided content in a MEMORY.md file, which is explicitly loaded at the start of every future session for the project. This allows malicious instructions to be stored and persistently influence the agent's behavior.
Ingestion points: The <what to remember> argument in the /si:remember command (SKILL.md).
Boundary markers: Absent; entries are appended as simple markdown list items without delimiters or protective instructions.
Capability inventory: The skill uses grep for searching (SKILL.md) and performs file append operations to $HOME/.claude/projects/.
Sanitization: No input validation, escaping, or filtering of the user-provided text is performed before storage.
[COMMAND_EXECUTION]: The skill utilizes shell command substitution and piping (pwd | sed, grep) to resolve local paths and search through existing memory files (SKILL.md). While these are standard utility commands, they represent execution of shell processes on the host system based on current directory state.

remember