Jamkris/everything-gemini-code:skills/content-hash-cache-pattern

The provided code snippets demonstrate file system operations that read from arbitrary file paths (`path` in `compute_file_hash`, `file_path` in `extract_with_cache`) and write to a configurable cache directory (`cache_dir` in `write_cache` and `extract_with_cache`). Specifically, `compute_file_hash` and the implicit `extract_text` function (called by `extract_with_cache`) read content from `file_path`. The `write_cache` function writes processed data to `cache_dir`. If an AI agent implements this pattern and allows untrusted input to control these parameters, it could lead to severe security vulnerabilities. An attacker could coerce the agent into reading sensitive files from the system (data exfiltration) or writing processed data to arbitrary locations, potentially overwriting critical system files, leading to denial of service, or even privilege escalation.