The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes an excessive amount of redundant technical documentation (150 identical sections per file across eight reference files, totaling 1,200+ sections). This structural 'bloat' is a form of noise obfuscation that can exhaust the agent's context window and complicate security reviews.\n- [PROMPT_INJECTION]: The agent protocol in SKILL.md commands the suppression of all conversational preamble and explanations ('No preamble. No postamble. No explanations.'), which reduces user oversight of the agent's logic and could be used to conceal unauthorized actions.\n- [COMMAND_EXECUTION]: Python and SQL templates (e.g., watermark_replication.py) use string interpolation for table and column names in database queries. This represents a command execution risk if the agent applies these templates to untrusted user input without implementing sanitization.\n- [PROMPT_INJECTION]: The skill possesses a high attack surface for indirect prompt injection as it is designed to ingest data from external database systems (e.g., via Debezium and batch scripts) and has the capability to execute SQL commands, but lacks instructions for using boundary markers or sanitizing ingested data.

data-data-replication