skills/erphq/skills/auth-clerk/Gen Agent Trust Hub

auth-clerk

Pass

Audited by Gen Agent Trust Hub on Jul 6, 2026

Risk Level: SAFEDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill accesses sensitive local credentials stored at ~/.build-host/credentials.json. This is standard behavior for the build.host vendor tool to manage authentication with its own API.
  • [PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks as it instructs the agent to ingest and analyze untrusted data from the existing project codebase.
  • Ingestion points: The agent is instructed to "inspect the existing stack" and operating on the project directory or git repository.
  • Boundary markers: The skill includes explicit "Safety Rules" that instruct the agent to confirm destructive actions and never print API keys or secrets.
  • Capability inventory: The agent has the capability to perform local repository edits and interact with the build.host API (PATCH envs, POST deploy, GET logs).
  • Sanitization: The skill lacks explicit sanitization instructions for ingested code, but mandates verification of the final deployment URL.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 6, 2026, 04:43 PM
Security Audit — agent-trust-hub — auth-clerk