The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from multiple sources to automate CRM updates and campaign generation, creating a surface for indirect prompt injection.
Ingestion points: Lead conversation threads (LinkedIn/Email) retrieved via the La Growth Machine MCP, website content discovered through sitemap crawling, and documents from Notion or Google Drive.
Boundary markers: Absent. The instructions do not define specific delimiters or "ignore" directives for instructions that might be embedded within lead replies or website content.
Capability inventory: The skill possesses the ability to execute shell commands (curl, python3), write to the local /tmp/ directory, and perform authenticated writes to the CRM (updating leads and campaigns via MCP).
Sanitization: The engine (scripts/build.py) performs automated validation of generated sentences and HTML payloads (e.g., regex checks for malformed URLs, length constraints, and variable schema validation) before they are written to the CRM.
[DYNAMIC_EXECUTION]: The helper script uses dynamic module loading to verify file existence.
Evidence: scripts/build.py uses __import__("os") to check for the presence of the library index file during execution.
[COMMAND_EXECUTION]: The skill workflow relies on executing shell commands for network resolution and internal processing logic.
Evidence: SKILL.md instructs the agent to use curl -sI to resolve short links and invokes python3 scripts/build.py for sitemap discovery, indexing, and matching tasks.

outbound-nurture-engine