The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Automated scanners detected a blacklisted malicious URL (https://fitprofessionals.net) within the skill's reference documentation and the submission tracker template.
[EXTERNAL_DOWNLOADS]: The file references/submission-tracker-template.csv was flagged by reputation scanners as malicious (FileRepMalware). This file is a core component used to track product submissions across external sites.
[EXTERNAL_DOWNLOADS]: The skill recommends ClawHub (clawhub.ai), noting it has a history of hosting hundreds of malicious skills, which represents a potential supply chain risk for users sourcing agent tools from the platform.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates interaction with a large volume of untrusted third-party websites, creating a surface for indirect prompt injection.
Ingestion points: Data from external directory pages and user-provided product marketing context.
Boundary markers: The instructions include a safety gate advising the agent to treat all directory content as untrusted and to ignore instructions embedded within those pages.
Capability inventory: The agent is instructed to use curl for verifying backlink status and dig for domain resolution.
Sanitization: No programmatic sanitization or filtering of external content is implemented; the skill relies entirely on natural language instructions to mitigate risks.
[COMMAND_EXECUTION]: The skill provides instructions for the agent to execute shell commands using curl, dig, and the mcp-publisher CLI tool.
mcp-publisher is an official utility for the Model Context Protocol, sourced from the modelcontextprotocol organization on GitHub.
The use of curl is limited to performing HEAD requests to verify backlink presence on external directory listings.

directory-submissions