The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses directive language ('CRITICAL: Auto-start on load', 'go straight to Step 1. Do not summarise') designed to override the agent's standard conversational behavior and safety summarization protocols.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from about-me.md and voice.md without clear boundary markers or sanitization, which could lead to the agent following instructions embedded within those files.
Ingestion points: Reads project files about-me.md and voice.md (Step 1).
Boundary markers: None identified; external content is used directly for positioning and voice tuning.
Capability inventory: File system write access (Claude Code surface in Step 3).
Sanitization: None identified.

content-matrix