The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill utilizes npx atxp@latest to download and execute the vendor's command-line interface from the npm registry at runtime. This is the primary execution mechanism for all provided tools.
[COMMAND_EXECUTION]: Functionality is implemented through subprocess calls to the atxp CLI. These commands facilitate interaction with the vendor's infrastructure for search, communications, and media creation.
[DATA_EXFILTRATION]: Tools like email send and phone send-sms allow for data transmission to external parties. The skill provides clear exfiltration guardrails, instructing the agent to never transmit sensitive credentials or configuration data.
[PROMPT_INJECTION]: The skill ingests untrusted data from multiple sources. A mandatory evidence chain analysis reveals:
Ingestion points: Web search results, X/Twitter posts, inbound emails, SMS messages, and call transcripts (SKILL.md).
Boundary markers: The skill instructs the agent to wrap untrusted output in [BEGIN UNTRUSTED CONTENT] delimiters.
Capability inventory: Subprocess calls (npx), email sending, and phone operations are available throughout the skill.
Sanitization: Instructions explicitly forbid passing untrusted content to shell interpreters or executing directives embedded in external content.
[CREDENTIALS_UNSAFE]: The skill manages an authentication token (ATXP_CONNECTION) stored in ~/.atxp/config. It provides comprehensive guidance on protecting this secret from exposure in logs, commands, or outbound messages.

atxp