projects
Fail
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: HIGHCREDENTIALS_UNSAFECOMMAND_EXECUTION
Full Analysis
- [CREDENTIALS_UNSAFE] (HIGH): The script 'scripts/railway-api.sh' reads the Railway authentication token directly from the user's home directory at '~/.railway/config.json'. This grants the AI agent full administrative access to the user's Railway account and projects.
- [COMMAND_EXECUTION] (MEDIUM): The skill utilizes bash heredocs and dynamic command construction in 'SKILL.md' to pass queries and variables to the 'railway-api.sh' helper. This pattern can be exploited if inputs are not correctly sanitized by the agent before execution.
- [INDIRECT_PROMPT_INJECTION] (LOW): The skill ingests untrusted data from the Railway API through 'railway list' and 'railway whoami' commands, which create a surface for indirect prompt injection from project or workspace metadata.
- [Indirect Prompt Injection Evidence]: 1. Ingestion points: 'SKILL.md' (railway list, railway whoami). 2. Boundary markers: Absent. 3. Capability inventory: Subprocess calls to 'railway' CLI and 'curl' via 'scripts/railway-api.sh'. 4. Sanitization: Absent.
Recommendations
- AI detected serious security threats
Audit Metadata