The Agent Skills Directory

[CREDENTIALS_UNSAFE]: Hardcoded password for a test identity found in configuration data.\n
File: config/identity.json\n
Evidence: The paypal.password field contains the value or#.2w}K. While this is associated with a mock 'Lex Luthor' identity on the vendor's domain (navins.biz), hardcoding credentials in configuration files is a security risk if the data were real.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its automated codebase reconnaissance phase.\n
Ingestion Points: The agent is instructed to scan routes/, app/Models/, src/types/, and tests/ directories on the user's filesystem to extract logic and constraints (File: SKILL.md).\n
Vulnerability: The 'Recon First' instruction explicitly states 'Codebase truth TRUMPS assumptions' (File: SKILL.md). This mandates the agent to prioritize information found in the code, creating a surface where malicious instructions embedded in comments or strings within the analyzed code could influence the agent's behavior during execution.\n
Capability Inventory: The agent has browser:navigate, browser:interact, and filesystem:read permissions (File: SKILL.md), enabling it to act on instructions read from the code.\n
Sanitization: No sanitization, boundary markers, or instructions to ignore embedded prompts are provided for the content read from the codebase.

ask-smart-booking-test