The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes documentation from local markdown files and fetches content from an external official domain, which creates a potential surface for indirect prompt injection.
Ingestion points: The skill ingests data from local files in the references/ directory and fetches live documentation pages from https://qwenwork.cn/docs/.
Boundary markers: The skill includes explicit instructions in SKILL.md for the agent to treat documentation content (including text, screenshots, and examples) as informational data only and not as executable instructions for the current task.
Capability inventory: The skill utilizes a local Python script (scripts/search_docs.py) for indexing and retrieving documentation snippets. No dangerous capabilities like arbitrary remote code execution or unauthorized network writing were identified.
Sanitization: The instructions mandate that the agent must verify specific user authorization before performing any state-changing actions (such as deletions or configuration changes) that might be described in the documentation.

qwenwork-wiki