The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands, specifically sed and grep, to programmatically determine the path to the agent's project-specific memory directory and search its contents for keywords.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from user-provided descriptions and project-specific memory files to generate new skill files and instructions. * Ingestion points: Data enters the system via the /si:extract command and by reading files in the $HOME/.claude/projects/ directory. * Boundary markers: There are no identified delimiters or explicit instructions to ignore embedded prompts within the ingested memory data. * Capability inventory: The skill possesses capabilities for file system reading (via grep) and file system writing (via the skill-extractor agent). * Sanitization: The skill instructions do not specify any validation or sanitization of the ingested data before it is used to generate the new SKILL.md and README.md files.

extract