The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data which influences subsequent file-writing and network actions.
Ingestion points: The user message is processed across multiple logic steps, including preferences, project_fact_ledger, outline_steps, and deliver_project_pack in SKILL.md.
Boundary markers: Prompts utilize explicit textual delimiters such as "User request:" and "Clarification:" to separate data from instructions.
Capability inventory: The skill has the ability to append data to the filesystem via memory_save (targeting memory/meta-kid-projects.md) and perform network queries via search and weather tools.
Sanitization: User-supplied input is escaped using the xml_escape filter prior to being interpolated into instructions.
[SAFE]: The skill implements multi-layer safety classification steps specifically designed to detect and block unsafe or age-inappropriate project requests.
[SAFE]: Filesystem and network capabilities are employed for their intended functional purposes, such as persisting user preferences and researching project materials.

meta-kid-project-planner