The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains instructions to 'Ignore Claude-specific mode-switch instructions', which is a pattern used to bypass platform-level constraints. It also enforces a 'Strict execution contract' to prioritize its own protocol over platform guidelines.
[COMMAND_EXECUTION]: The skill explicitly instructs the agent to use shell commands such as git status and git diff, and to execute a repository-local Python script .claude/scripts/code_graph for analysis and impact tracing.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted frontend source files (SCSS, TS, HTML) for review. While it employs an 'Evidence-Based Reasoning' framework, it lacks specific technical boundary markers or sanitization for these ingestion points, representing a potential surface for indirect injection.

ui-review