The Agent Skills Directory

[PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it ingests and processes content from local files, directories, and task artifacts.\n
Ingestion points: Data enters the system through file, directory, and task modes, which read contents from various filesystem paths (documented in SKILL.md).\n
Boundary markers: The system uses Markdown for storage but lacks explicit boundary markers or 'ignore' instructions for the LLM when it later retrieves and processes these memories.\n
Capability inventory: The skill can execute shell commands via Bash, write or modify files, delete files using rm, and perform git commits.\n
Sanitization: There is no mention of content sanitization or filtering to prevent the inclusion of malicious instructions in the memory vault.\n- [COMMAND_EXECUTION]: The skill uses shell commands to perform vault maintenance and search operations.\n
Evidence: SKILL.md defines workflows using grep, wc, bc, ls, git, and rm. The gc sub-mode specifically performs permanent file deletion.\n
Control: The skill includes mandatory interactive requirements, explicitly instructing the agent to use AskUserQuestion and wait for confirmation before any write or delete operations are performed.

skill-memory