The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: Automated scanner flagged installation instructions in the SKILL.md file as a potential threat. Analysis confirms this is the official Railway CLI installation command (bash <(curl -fsSL cli.new)) provided as documentation for the user. No script within the skill executes remote code or downloads untrusted content for autonomous execution.
[PROMPT_INJECTION]: Automated flags for action concealment in database analysis references are false positives. They describe instructions for the agent to exclude internal platform metrics (resource ceilings on auto-scaling instances) that could be misleading to the user. This is a legitimate UI formatting rule and not a malicious attempt to hide behavior.
[DATA_EXFILTRATION]: All network operations, including API requests and log retrieval, target verified Railway infrastructure domains such as backboard.railway.com and station-server.railway.com. No credentials or sensitive data are transmitted to third-party or unknown services.
[COMMAND_EXECUTION]: The skill executes shell commands and database queries via local Python and Bash scripts. Commands are constructed from hardcoded templates or resource IDs extracted from Railway URLs. SQL queries are Base64-encoded before transmission to ensure safe execution over SSH.
[INDIRECT_PROMPT_INJECTION]: The skill processes database logs, which are considered untrusted external data. This risk is mitigated by explicit instructions that prioritize analysis over state-changing actions and require direct user confirmation for any mutations.

use-railway