The Agent Skills Directory

[PROMPT_INJECTION]: The skill employs extremely forceful and authoritative language (e.g., 'ABSOLUTELY MUST', 'NOT NEGOTIABLE', 'YOU DO NOT HAVE A CHOICE') to override the agent's internal autonomy. It specifically commands the agent to ignore its own logical reasoning processes—such as the need for context or preliminary exploration—labeling them as 'Red Flags' to be disregarded in favor of immediate tool invocation.
[INDIRECT_PROMPT_INJECTION]: The instructions mandate that the agent must load external content via the Skill tool and follow it 'directly' and 'exactly'. This creates a vulnerability surface where potentially untrusted external instructions are given authoritative priority over the agent's native safety protocols and logic.
Ingestion points: SKILL.md (instructions forcing tool usage).
Boundary markers: Absent; the skill explicitly directs the agent to follow external content without question.
Capability inventory: Skill tool (read/load capabilities) and TodoWrite tool (write capability).
Sanitization: None; instructions compel the agent to adopt external skill content as mandatory constraints.

using-ltk