The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes and transforms user-supplied text to match a specific style, which creates a potential surface for instructions embedded within the input to influence the agent's execution.
Ingestion points: User text submitted for drafting, editing, or style application via SKILL.md.
Boundary markers: The skill employs a two-pass workflow and a mandatory checklist of 'hard floors' (§17 and §18) to enforce factual accuracy and override potential overrides in the input.
Capability inventory: The skill uses memory_search, memory_save, and ask_user_question to manage context and gather facts.
Sanitization: The skill uses an internal verification step (Pass 2) to audit the draft against safety and style requirements.
[SAFE]: The skill references established and well-known services, including the author's GitHub profile and Mermaid.live for diagram rendering. These are used for attribution and documentation purposes.
[SAFE]: The use of internal memory tools (memory_save and memory_search) follows best practices for maintaining session consistency and factual grounding, with instructions to avoid storing sensitive or non-verified data.

writing-style