The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a local Python script (.codestable/tools/search-yaml.py) via the shell to find existing documents.
The commands described in the '搜索工具' (Search Tools) section and Phase 1.5 use placeholders like {组件名} (component name) and {主要 tag} (main tag).
If these placeholders are filled with unsanitized user input or content from external files, it could lead to shell command injection (e.g., injecting ; rm -rf / or similar payloads).
[PROMPT_INJECTION]: The skill involves reading existing learning documents from the .codestable/compound/ directory to perform updates (Phase 1.5).
The instructions do not include the use of boundary markers or specific warnings to ignore instructions that might be embedded within the documents being read, making the skill susceptible to indirect prompt injection.
[DATA_EXPOSURE]: The skill specifically targets project metadata and configuration files, such as .codestable/attention.md and shared-conventions.md.
While this is the intended functionality of the skill, it involves the agent reading and potentially acting upon internal project documentation and constraints.

cs-learn