The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The installation instructions direct the user to clone the dashboard source code from a third-party repository at https://github.com/builderz-labs/marketing-dashboard.git.
[REMOTE_CODE_EXECUTION]: The skill requires the execution of setup and maintenance scripts, such as pnpm env:bootstrap and ./scripts/template-audit.sh, which are contained within the external repository. Executing scripts from a downloaded project involves a risk of arbitrary code execution.
[COMMAND_EXECUTION]: The skill utilizes several shell commands for environment setup, building, and database management, including git, pnpm, and local shell scripts.
[INDIRECT_PROMPT_INJECTION]: The application creates an attack surface by ingesting untrusted data that may later be processed by AI agents.
Ingestion points: Data provided to the lead creation API (POST /api/crm/leads) and agent metadata read from the filesystem (agent.json).
Boundary markers: The code snippets do not show explicit delimiters or "ignore embedded instructions" warnings for data that might be passed to LLM agents.
Capability inventory: The skill has the capability to perform SQLite database writes (better-sqlite3), enroll leads in outreach sequences, and read configuration files from the disk.
Sanitization: The implementation uses parameterized SQL queries to prevent technical injection, but it does not demonstrate content sanitization or filtering to prevent instruction-based injection from lead metadata or agent descriptions.
[DYNAMIC_EXECUTION]: The skill performs runtime discovery of AI agents by scanning directories and reading configuration files based on paths constructed from environment variables (HERMES_OPENCLAW_HOME). It also dynamically parses instance configurations from the HERMES_OPENCLAW_INSTANCES JSON environment variable.

hermes-marketing-dashboard