The Agent Skills Directory

[PROMPT_INJECTION]: The file memoryskill.md contains persistent instructions that dictate the agent's behavior and constraints, such as mandatory output formats ("必须包含 Word 文档 (.docx)", "严禁只交付 Markdown") and specific role-playing requirements ("数学建模专家").
[PROMPT_INJECTION]: Indirect prompt injection surface detected.
Ingestion points: memoryskill.md (specifically the "External Resources / Literature" section in the Short-term Workbench) is designed to store data retrieved from external sources like g-sci/authoritative-data-harvester.
Boundary markers: Absent; the skill lacks specific delimiters or instructions to ignore commands embedded within the stored data.
Capability inventory: No executable code is provided within this skill; however, the agent is instructed to use external scripts in the skills/ directory.
Sanitization: Absent; there is no logic defined to sanitize or validate the content ingested from external literature or data sources.
[NO_CODE]: The skill consists entirely of Markdown files for documentation and memory storage. It does not contain any executable scripts (Python, JavaScript, etc.), though it references the use of scripts located in other skill directories.

context-memory-keeper