The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions in SKILL.md direct the agent to execute a provided Python script (scripts/atoms-search.py) using python3 to retrieve knowledge base entries based on user queries.
[DATA_EXFILTRATION]: The atoms-search.py script performs network operations to https://api.xyskill.xyz via HTTP POST requests. It transmits search keywords derived from user input to this external domain. While consistent with the skill's business purpose and associated with the vendor's naming patterns, the domain is not on the trusted services list.
[INDIRECT_PROMPT_INJECTION]: The skill incorporates untrusted user input into its logic flow, creating an attack surface where user-provided text influences command-line arguments and network payloads.
Ingestion points: User status updates and startup questions processed in Phase 1 and Phase 2.
Boundary markers: None identified; user input is passed directly to the keyword extraction logic.
Capability inventory: Subprocess execution for script running and network POST operations to a remote API.
Sanitization: The Python script uses regular expressions to tokenize keywords but lacks rigorous escaping or validation of the resulting data before its use in shell commands or HTTP requests.
[DATA_EXPOSURE]: The script atoms-search.py probes multiple filesystem paths, including common AI agent directories in the user's home folder (e.g., ~/.claude/skills/xy, ~/.agents/skills/xy), to locate its local JSONL database.

xy-newbie