The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface through its 'unconfirmed items' (待确认事项) processing logic. It is designed to scan project documentation for specific headers and extract list items to be processed as instructions or decision points.
Ingestion points: The skill reads from multiple stage-specific documents including '01-需求/需求文档.md', '02-设计稿/设计稿.md', '03-需求分析/需求分析文档.md', '04-技术方案/技术方案.md', and '05-实现清单/实现清单.md'.
Boundary markers: The skill relies on Markdown headers such as '## 待确认项', '## 待确认事项', or '## 待解决问题' and list item markers ('^- ') to delimit untrusted content.
Capability inventory: The skill has the ability to update the central '.dwf/state.json', modify individual '_meta.json' files, and write user decisions back into the source documentation files. While it does not directly execute code, it controls the flow of the entire development agent.
Sanitization: No explicit sanitization or instruction-filtering of the content extracted from these documents is mentioned in the rules.
[NO_CODE]: The provided skill consists entirely of instructional Markdown and JSON configuration files. No executable scripts (Python, JavaScript, shell, etc.) were found in the skill package.

dwf-orchestrator