The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes arbitrary codebase files (via Read, Grep, Glob, and Task tools) to identify entities, relationships, and components. These source files are untrusted and could contain malicious instructions (e.g., hidden in code comments) designed to influence the agent's behavior during analysis or diagram generation.
Ingestion points: Any file within the target repository accessed during the exploration and parsing steps (SKILL.md).
Boundary markers: Absent. No instructions are provided to ignore or escape potentially malicious content within the source files.
Capability inventory: The skill can Write to the docs/ directory and spawn subagents via the Task tool.
Sanitization: Absent. Content extracted from files is interpolated directly into templates in assets/templates/.
[COMMAND_EXECUTION]: The references/detection-patterns.md file contains shell command snippets (e.g., find . -name "*models.py" -exec grep -l ...) which the skill instructs the agent to use for codebase analysis. If interpreted as raw shell commands in an environment that provides a shell tool, this could potentially be exploited by maliciously named files or paths.

docs-diagram