The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources, including prospect lists (gtm/send-list.csv) and email replies (gtm/replies.jsonl). \n- Ingestion points: Prospect data (CSV) and prospect replies (JSONL) are read into the agent's context for scoring and triage. \n- Boundary markers: The skill mandates strict formatting rules and uses specific delimiters (e.g., "Quotation marks only around words the source itself quotes") to separate signal data from agent instructions. \n- Capability inventory: The skill can execute local scripts via Bash, perform network lookups via dig/host, and read/write files within the project directory. \n- Sanitization: Input data is passed through deterministic Python scripts (score_letter.py, score_list.py, score_reply.py) that validate content against predefined lexicons and rules before processing. \n- [COMMAND_EXECUTION]: The skill invokes local Python and Bash scripts to perform domain health checks and content scoring. \n- The allowed-tools configuration limits execution to a specific set of scripts located in ${CLAUDE_PLUGIN_ROOT}/scripts/. \n- Tools like dig, host, and nslookup are used for DNS deliverability audits.

cold-email