review
Warn
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill is instructed to fetch a remote file (https://docs.aevral.com/agent-setup/prompt.md) and follow its instructions. This dynamic instruction loading allows external content to influence the agent's behavior at runtime.
- [INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection via the remote prompt file. * Ingestion points: Remote instruction fetch from docs.aevral.com. * Boundary markers: Present; the skill defines a 'Boundary' section in SKILL.md that takes precedence over fetched content. * Capability inventory: Shell command execution via the GitHub CLI and network requests to the Aevral API. * Sanitization: Absent; the skill does not verify or sanitize the remote content before processing.
- [COMMAND_EXECUTION]: The skill uses the GitHub CLI (
gh api) to retrieve repository identifiers for use in API payloads. - [EXTERNAL_DOWNLOADS]: The skill communicates with external endpoints at aevral-worker-prod.fly.dev and aevral.com to manage billing, usage, and security scans.
Audit Metadata