The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from LinkedIn profiles and social posts via Crustdata APIs and writes this information into local configuration files (config/persona-profile.md and config/gtm-config.md). These files are then consumed by other GTM-related skills, creating a potential vector for indirect prompt injection if the source data contains malicious instructions.
Ingestion points: External data from LinkedIn profiles and posts fetched via the person_enrich and social_post_list_live tools.
Boundary markers: The skill uses markdown structure for the configuration files but does not explicitly mention escaping logic for the retrieved content.
Capability inventory: The skill can execute code, call remote tools, and write files to the local filesystem.
Sanitization: Relies on a mandatory human-in-the-loop review step where the agent must show the persona to the user for verification before writing files.
[COMMAND_EXECUTION]: The skill uses the execute tool to run JavaScript code that orchestrates tool calls and processes data. The provided code snippets are limited to legitimate business logic for data enrichment.
[EXTERNAL_DOWNLOADS]: The skill references the author's own infrastructure at crustdata.com and install.crustdata.com for data enrichment and MCP server installation. These references are consistent with the skill's functionality and use the vendor's official domains.

icp-builder