The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted content from multiple external sources, including YouTube comments, Discord messages, Crisp chats, and CRM notes.
Ingestion points: Data enters the agent context through tools like mcp__cogny__mcp__YouTube__list_comment_threads, mcp__cogny__mcp__Discord__get_channel_messages, and CRM search tools specified in SKILL.md.
Boundary markers: The skill uses a structured 'Failures: none' footer and relies on human review via tickets/findings for all output, but does not use explicit delimiters for external content within scripts.
Capability inventory: The skill has access to Bash, Write, Edit, and platform-specific tools for creating automations, tickets, and scripts.
Sanitization: The instructions recommend using jq for JSON filtering and deduplication, but lack specific natural language sanitization for embedded instructions in comments or chats.
[COMMAND_EXECUTION]: The Bash tool is explicitly allowed and utilized for operational tasks, such as fetching page metadata with curl and processing JSON data with jq as described in the references/mcp-routing.md file.
[DYNAMIC_EXECUTION]: The skill utilizes the Cogny platform's create_script capability to generate deterministic JavaScript rules for marketing spend caps and logic thresholds.
[EXTERNAL_DOWNLOADS]: The skill connects to the vendor's official service at https://app.cogny.com/mcp for its primary data operations. This resource is provided by the skill author (Cogny AI).

marketing-engineer