The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied text to build writing profiles and perform rewrites. \n- Ingestion points: Untrusted content enters the agent context when the user provides text to be rewritten or shares writing samples from external sources like email or Slack. \n- Boundary markers: The instructions lack specific delimiters or safety markers to isolate user input from the skill's system instructions. \n- Capability inventory: The skill performs file read and write operations on the local file system at ~/.claude/ghostwriter-profile.md and invokes external tool-based skills (check and polish). \n- Sanitization: No evidence of sanitization, escaping, or validation of external content is performed before the data is processed.

ghost