The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a bundled Python script (scripts/atoms-search.py) to perform search operations within its knowledge base.
Evidence: The Phase 2 instructions in SKILL.md provide a specific command line for the agent to run: python3 <本 skill 目录>/scripts/atoms-search.py "共鸣情绪立场转发" -k 5.
[EXTERNAL_DOWNLOADS]: The helper script initiates network connections to a remote API to retrieve data when local files are insufficient or when instructed to use the cloud database.
Evidence: The script scripts/atoms-search.py contains logic to perform POST requests to https://api.xyskill.xyz/v1/atoms/search using the urllib.request library.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted content provided by users (such as drafts, scripts, or viral content), which could contain hidden instructions intended to manipulate the agent's behavior.
Ingestion points: The skill processes user-supplied text during Phase 1 (Diagnosis/Decoding) as defined in SKILL.md.
Boundary markers: The instructions do not define specific delimiters or "ignore instructions" wrappers to protect the agent from embedded commands in the user's content.
Capability inventory: The agent has the capability to execute a Python script that accesses the network and reads local files (including searching through other skill directories like ~/.claude/skills/).
Sanitization: The input text is passed to analysis logic without explicit sanitization or filtering for prompt injection patterns.

xy-echo-test