The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted user-provided code and requirements, which exposes it to potential indirect prompt injection attacks. • Ingestion points: User-supplied 'code or requirements' provided by the user, as specified in SKILL.md and the default prompt in agents/openai.yaml. • Boundary markers: Instructions lack explicit delimiters or safety warnings to ignore instructions embedded within the user-provided content. • Capability inventory: The agent utilizes the 'astrale call' CLI for deployment, secret management, and deletion, along with 'npm exec' and 'npm install' for building artifacts (documented in references/workflows.md and references/service-account-cron.md). • Sanitization: No specific sanitization or validation protocols for external content are described.
[COMMAND_EXECUTION]: The skill provides detailed workflows for executing platform-specific CLI commands to manage service infrastructure. • Evidence: Instructions include 'astrale call' for deploying workers, managing write-only secrets, and convergent deletion (references/workflows.md). • Evidence: The skill uses 'npm exec' to run project scaffolding tools and 'npm install' for dependency management (references/service-account-cron.md).
[EXTERNAL_DOWNLOADS]: The skill references and executes external packages and code repositories to fulfill its purpose. • Evidence: Downloads vendor-owned packages including '@astrale-os/adapter-astrale', '@astrale-os/sdk', and '@astrale-domains/services'. • Evidence: Utilizes the 'jose' library for cryptographic operations and references the 'astrale-os/domains' GitHub repository for scaffolding logic (references/service-account-cron.md).

astrale-services