The Agent Skills Directory

[PROMPT_INJECTION]: The template file assets/general/agent-skill-architecture.md contains directive language and role-play instructions ('Role: You are an expert AI Engineer', 'Now, await the user's request. Ask them... GO:') designed to override the agent's behavior and take control of the interaction context.
[DYNAMIC_CONTEXT_INJECTION]: The skill documents and provides examples for using the shell-injection syntax (![ command ]) in its agent-skill-architecture.md template, specifically mentioning ![ pwd ], ![ find . ], and ![ cat package.json ] as mandatory injections for coding skills.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to restructure user-provided documents ('argumentHint: the document to draft or restructure'). It lacks input sanitization or boundary markers to prevent the agent from executing instructions potentially hidden within those documents. 1. Ingestion points: User-supplied documents passed as arguments to the skill (SKILL.md). 2. Boundary markers: No explicit delimiters or 'ignore instructions' warnings are provided in the prompts. 3. Capability inventory: The agent has file system read and write capabilities, and the templates suggest the use of shell execution via dynamic context injection. 4. Sanitization: No evidence of escaping or filtering external content before processing.

write-document