The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs local code analysis using standard utilities like find, grep, rg, and sed. These operations are limited to identifying Python files and extracting import statements within the local project directory for architectural mapping.
[DYNAMIC_EXECUTION]: The skill locates and executes a Python script from the local plugin directory ~/.claude/plugins. This is the intended behavior for extending functionality through local plugins and is constrained to a specific path.
[INDIRECT_PROMPT_INJECTION]: The skill ingests source code from the repository, which creates an inherent surface for indirect prompt injection if files contain adversarial content designed to mislead the agent. * Ingestion points: Source code files read by standard grep and ripgrep commands (SKILL.md). * Boundary markers: No explicit delimiters are used for the code content during ingestion. * Capability inventory: Local shell command execution and script invocation (SKILL.md). * Sanitization: None; the raw results of the analysis are passed to the agent for interpretation and visualization.

code-communities