The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill includes a confirmed malicious URL (https://fitprofessionals.net) in its recommended directory lists within references/directory-list.md and the tracking template. Interactions with known malicious domains pose a significant risk of malware exposure.
[REMOTE_CODE_EXECUTION]: The file references/submission-tracker-template.csv was flagged by automated scanners with a high-confidence malware detection (FileRepMalware). Malicious CSV files are frequently used to deliver payloads via CSV Injection or DDE attacks.
[COMMAND_EXECUTION]: The instructions in SKILL.md direct the agent to use curl -fsSL to fetch the raw HTML content of user listings from arbitrary external directories. Executing shell commands against untrusted remote content can lead to command injection or system compromise if the output is not strictly sanitized.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external directory websites. These pages are attacker-controllable surfaces that may contain hidden instructions intended to override the agent's behavior or exfiltrate sensitive context. While the skill includes a self-identified safety warning, the capability to read and act on external data remains a persistent vulnerability surface.
Ingestion points: Data is pulled from external URLs via curl as specified in the 'Verify the Actual Backlink' section of SKILL.md.
Boundary markers: The skill mentions a 'safety gate' warning the agent to treat directory pages as untrusted data, but does not implement robust structural delimiters.
Capability inventory: The agent has the capability to execute shell commands (curl, dig, mcp-publisher) and interact with the local filesystem (reading/writing the CSV tracker).
Sanitization: No explicit automated sanitization or escaping of the fetched HTML content is implemented before processing.

directory-submissions