The Agent Skills Directory

[DATA_EXFILTRATION]: The skill includes a script that initiates network communication with an external, non-whitelisted domain belonging to the vendor.
Evidence: The script scripts/atoms-search.py uses urllib.request.urlopen to perform POST requests to https://api.xyskill.xyz/v1/atoms/search.
Context: The script transmits the search query and skill name to this endpoint to retrieve relevant knowledge base entries. As the domain is vendor-owned, this represents standard functionality but involves transmitting user-influenced search terms to an external server.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data, creating an attack surface for indirect prompt injection.
Ingestion points: The skill processes video links, local files, and transcripts provided by the user or external sources during the Phase 1 "收对标" step in SKILL.md.
Boundary markers: The instructions mandate "停顿规则" (pausing rules) to prevent continuous automated execution and "证据清点" (evidence grading) to explicitly separate observed facts from inferences.
Capability inventory: The skill utilizes a local Python script scripts/atoms-search.py to retrieve structured knowledge based on the user's input.
Sanitization: The skill implements "卖点门控" (sell point gating) which prevents the generation of scripts if provided product selling points are vague or missing, mitigating the risk of AI-fabricated content.
[COMMAND_EXECUTION]: The skill instructions involve the execution of a locally provided script for data lookup.
Evidence: SKILL.md instructs the agent to execute python3 skills/xy-replica/scripts/atoms-search.py with various flags for knowledge retrieval. This is a legitimate tool invocation within the skill's defined scope.

xy-replica