The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (source code, logs, and git diffs) through tools like view_file and grep_search to generate human-readable explanations.
Ingestion points: The skill uses view_file, list_dir, grep_search, and git diff to read content from the local codebase (as specified in the execution workflow in SKILL.md).
Boundary markers: The instructions lack explicit delimiters or "ignore embedded instructions" warnings to separate untrusted code content from agent instructions, potentially allowing adversarial code comments to influence agent behavior.
Capability inventory: The skill utilizes read-only tools such as view_file, list_dir, grep_search, and git diff for codebase exploration (SKILL.md).
Sanitization: No specific sanitization, filtering, or validation logic is provided for the ingested source code or log data before it is processed for explanation.

tech-explainer