The Agent Skills Directory

[COMMAND_EXECUTION]: The skill workflow (SKILL.md step 4) includes running "构建或类型检查" (build or type checks) within the target project environment. This involves the agent executing shell commands to verify that the refactored CSS does not break the project's build process.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external data from the user's project files, which creates an attack surface for indirect prompt injection.
Ingestion points: The skill reads "指定样式与组件" (specified styles and components) from the target project (SKILL.md Workflow step 1).
Boundary markers: The skill does not explicitly define character-level boundary markers or delimiters for the ingested code, though it instructs the agent to separate analysis and preview from actual modification.
Capability inventory: The skill has the capability to read/write files and execute shell commands for build validation (SKILL.md Workflow steps 3 and 4).
Sanitization: There is no explicit evidence of automated sanitization or filtering of instructions that might be embedded in CSS comments or component metadata within the project files.
[PERSISTENCE]: The included scripts/profile_store.py script manages a local JSON configuration file (typically located at ~/.config/agent-skills/profile.json or ~/.lovstudio/skills/profile.json). This script allows the agent to read and persist user-stated values, brand information, and skill-specific records across different sessions. While intended for configuration management, it establishes a persistent state on the local filesystem.

lov-better-css