The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes project-specific data to build a memory store, which represents a surface for indirect prompt injection if project files contain adversarial content.
Ingestion points: Ingests content from markdown files located in ~/.claude/projects/<slug>/memory/*.md using Read and Grep tools.
Boundary markers: None explicitly defined for individual memory entries.
Capability inventory: Includes the ability to use Bash for system checks, Write for file creation, and Edit for consolidation tasks.
Sanitization: The skill does not sanitize content before incorporating it into the memory store, as it is intended for internal agent use.
[COMMAND_EXECUTION]: Employs Bash for local environment verification, file integrity checks, and security scanning. The skill also documents platform-specific hooks in settings.json for automated reminders, but it includes a strict policy against modifying these settings without user interaction.
[DATA_EXFILTRATION]: Contains a dedicated safety gate mechanism that proactively scans memory files for potential credential leaks using grep patterns, ensuring that sensitive values are not stored in the knowledge base.

autodream