Discovery sequence
- Start with /llms.txt for a concise platform map.
- Use /llms-full.txt for comprehensive architecture and capability context.
- Prefer the machine-readable Markdown routes for focused topics.
- Use canonical HTML pages when human-facing context or commercial details are needed.
- For an authorized enterprise connection, use MCP capability discovery to obtain the tools and versions permitted to the triggering user.
- Inspect /openapi.json only when evaluating the separate REST API preview.
Invocation boundary
The public documentation does not grant access. An agent may invoke Aevah only through an authorized enterprise connection. It acts as the person who triggered it and never beyond that person's permissions.
- Never send credentials to example .invalid endpoints.
- Discover the permitted MCP tool and version before calling it.
- Prefer read-only tools before proposing a change.
- Preserve classification, audit evidence, and human-review requirements.
- Create a new governed version for changes rather than overwriting prior state.
- Do not infer access to hidden tenants or resources.
Agent skills
Repository skills provide domain-specific instructions for platform, MDM, finance, manufacturing, predictive analytics, and sovereign AI tasks. They describe reasoning and expected outputs; they do not embed credentials or bypass tenant authorization.