The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides a submission tracker template (references/submission-tracker-template.csv) that includes a reference to https://fitprofessionals.net. This domain is identified as malicious and blacklisted by security scanners.
[EXTERNAL_DOWNLOADS]: The skill references the Linux Foundation's official Model Context Protocol registry on GitHub for technical integrations and Tier 4 registry submissions.
[COMMAND_EXECUTION]: The workflow in SKILL.md instructs the agent to execute shell commands using curl -sIL to verify the presence and status of backlinks on external directory sites. This facilitates automated network interaction with all domains listed in the reference files, including the flagged malicious domain.
[DATA_EXFILTRATION]: The skill's primary purpose is to guide users to submit product marketing data, founder stories, and asset URLs to external directories. By including a blacklisted domain in the recommended target list, the skill creates a high risk of sensitive project information being exfiltrated to a malicious actor.
[INDIRECT_PROMPT_INJECTION]: The skill processes lists of external domains and descriptions to generate submission plans, which serves as a vulnerability surface.
Ingestion points: The agent ingests directory data and URLs from references/directory-list.md and references/submission-tracker-template.csv to build the user's distribution plan.
Boundary markers: The skill lacks delimiters or protective instructions when reading these external lists to ensure the agent does not follow instructions potentially embedded in directory names or descriptions.
Capability inventory: The skill utilizes network capabilities (curl) to interact with the ingested URLs.
Sanitization: There is no evidence of validation or filtering for the URLs provided in the reference files, allowing malicious or blacklisted entries to be included in the agent's recommended actions.

directory-submissions