The Agent Skills Directory

Indirect Prompt Injection (LOW): The skill is designed to ingest untrusted data from the current conversation history and use it to update persistent files like CLAUDE.md and about-taylor.md.
Ingestion points: The skill explicitly reviews the entire conversation history for 'memory-worthy moments' and reads existing files such as .claude/memories/about-taylor.md and CLAUDE.md.
Boundary markers: There are no explicit instructions to treat extracted content as untrusted or to use delimiters when processing the conversation for new memories.
Capability inventory: The skill uses the Edit and Write tools to modify the agent's primary instruction file (CLAUDE.md) and profile (about-taylor.md).
Sanitization: No sanitization or validation of the extracted 'memories' is performed before they are proposed for inclusion in persistent storage, relying solely on human review at the presentation phase.
Command Execution (SAFE): The skill uses the Bash tool, but it is strictly restricted via the allowed-tools list to the date command (Bash(date:*)), posing no security risk.

debrief