The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process arbitrary user-provided files and directories to build a knowledge base.
Ingestion points: The scripts scripts/scan_materials.py and scripts/audit_knowledge_base.py are used to scan and inventory external file systems for the agent.
Boundary markers: The skill instructions in SKILL.md and references/material-intake.md explicitly require the agent to separate "原始事实" (original facts), "外部观点" (external opinions), and "用户判断" (user judgments) to prevent accidental obedience to instructions found in user documents.
Capability inventory: The agent has the capability to read local files, create directory structures, and write/update markdown files for logs, state reports, and knowledge base navigation.
Sanitization: The skill relies on instructional prompting boundaries and the audit script's sensitive file detection rather than strict technical content sanitization.

shengjiang-knowledge