The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data sources including git diffs, user-provided implementation files, and external requirement/plan documents to generate documentation in docs/spec/. This creates a vulnerability surface where malicious instructions embedded in source files could be persisted into the project's documentation.
Ingestion points: The skill reads git diff, user-specified files, requirements, and plans as outlined in the '工作流程' (Workflow) and '上下文读取口径' (Context Reading) sections.
Boundary markers: The instructions emphasize summarizing behavior rather than verbatim copying, which provides a logical filter, but there are no explicit instructions to use technical delimiters or escaping for the content written to the documentation files.
Capability inventory: The skill utilizes file system write operations to create or update files in docs/spec/ and executes version control commands (git diff, git status).
Sanitization: No explicit sanitization or filtering of input content is mandated beyond the instruction to "compress" information and avoid large code blocks.
[COMMAND_EXECUTION]: The workflow instructs the agent to execute shell commands for version control, specifically git diff and git status, to verify documentation changes. These are standard operations for development-oriented skills.

summary