The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze and summarize arbitrary codebase content, which inherently involves processing untrusted data that could contain malicious instructions.
Ingestion points: The skill uses tools like rg, find, and sed to read files from the repository during exploration and explanation steps as defined in the prompt templates.
Boundary markers: Prompt templates in references/explainer-prompt.md and references/explorer-prompt.md use markdown headers to structure instructions, though they do not include specific delimiters for ingested code content.
Capability inventory: The subagents are explicitly restricted to read-only access in SKILL.md, with instructions stating "inspect only, no writes, no git commands."
Sanitization: There is no evidence of sanitizing or escaping the content read from the codebase before it is provided to the subagents.
[COMMAND_EXECUTION]: The templates for subagents instruct the use of shell commands (rg, find, sed -n) to perform codebase searches and file reads. These operations are standard for the skill's primary purpose of architectural exploration and are performed under the restricted read-only constraints defined in the main instructions.

how