The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill utilizes a Python script (scripts/atoms-search.py) that performs network requests to https://api.xyskill.xyz/v1/atoms/search. This domain is used to fetch knowledge base entries (atoms) relevant to the diagnostic process.
[COMMAND_EXECUTION]: The skill invokes a local script, scripts/atoms-search.py, to search for and retrieve information from knowledge base files. This is a core part of its diagnostic functionality.
[DATA_EXFILTRATION]: The search script performs network operations to an external domain. While the intended use is to send search queries, the script also searches for local files in various host directories such as ~/.claude/skills/xy and ~/.agents/skills/xy. Although it does not directly exfiltrate system secrets, it involves network calls to a non-whitelisted domain and file access to user-specific application directories.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest user-supplied business data and combine it with external knowledge fetched from an API or local file, creating a surface for potential indirect prompt injection.
Ingestion points: User input gathered through multiple rounds of questioning (Phase 2) and external data retrieved via the search script (atoms.jsonl).
Boundary markers: The skill uses markdown tables and headers to structure its output, but it does not employ specific delimiters or instructions to ignore potential commands within the external data.
Capability inventory: The skill possesses the ability to execute a local Python script with network access and file read permissions.
Sanitization: There is no evidence of sanitization or filtering for the external knowledge content before it is presented to the user or used in the agent's reasoning.

xy-ip