jp-executor
Pass
Audited by Gen Agent Trust Hub on Jul 12, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill includes hard initialization gates, requiring pre-validated design documents (PRD and Architecture) before activation, which prevents arbitrary instruction execution through unverified inputs.
- [CREDENTIALS_UNSAFE]: The instructions explicitly forbid hardcoding secrets. It implements a secure credential handling flow where API keys are provisioned via authenticated tools and stored exclusively in local environment files (
.env), ensuring they never appear in logs or project artifacts. - [EXTERNAL_DOWNLOADS]: The skill fetches technical specifications and integration guides from official Juspay documentation domains (
*.juspay.io,*.juspay.in, and verified CloudFront CDNs). These operations are scoped strictly to the implementation context. - [COMMAND_EXECUTION]: Development tasks such as package installation and mobile SDK configuration are performed using standard package managers (npm, gradle, pod). The skill includes mandatory user checkpoints before executing risky commands, such as database schema changes or server tests.
- [DATA_EXFILTRATION]: Network activity is restricted to authenticated merchant platform operations and public documentation retrieval. There are no patterns suggesting unauthorized data collection or exfiltration to third-party servers.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted project documentation (
prd.md,architecture.md) to drive implementation. While this is an inherent injection surface for agentic tools, the skill mitigates this risk by requiring manual confirmation for all external or destructive actions and by grounding every code decision in re-verified documentation.
Audit Metadata