The Agent Skills Directory

[PROMPT_INJECTION]: Indirect injection risk via processed memory files. The skill reads and analyzes markdown files from the ~/.claude/projects/ directory, which may contain untrusted content from various project sources.
Ingestion points: Memory files (user_*.md, feedback_*.md, project_*.md, reference_*.md) and the MEMORY.md index file.
Boundary markers: None. The skill does not use delimiters or instructions to ignore potential commands embedded within the memory files it audits.
Capability inventory: The skill uses Bash, Write, and Edit tools to modify or delete files in the internal configuration directory.
Sanitization: None. The content is parsed for taxonomy markers (like 'Why:' and 'How to apply:') without escaping or validation.
[COMMAND_EXECUTION]: Access to and modification of sensitive internal configuration. The skill operates within the ~/.claude/ directory, which stores the agent's persistent memory and project context. Using the --prune flag allows for the deletion and modification of this data, which can alter the agent's behavior and available context in future sessions.

memory-curate