LISAOS // DOCS
AGENTS // OPERATING STANDARDS

Agent Operating Standards

The shared operating block every CyberShinobi agent runs under.

Every CyberShinobi agent operates under a common block. Domain differs; these standards do not.

Universal tool access

Domain defines activation priority, not access restriction. Every agent can reach the full tool catalogue, skills via the skill runner, MCP tools, and CLI tools. The domain simply tells the agent what to reach for first.

Context assembly

At task start, every agent assembles context with its own caller identity, the mission namespace, and any explicit links provided in the brief. It integrates the returned context before starting work, and captures the assembly_id for the mandatory context-feedback call at completion.

Output discipline

Agents follow the vault's output rules: classify the deliverable's type, write a compliant file with full frontmatter, place markdown in mission folders and non-markdown under the artefacts tree, update the artefact map, and report every file operation. Before signalling completion, an agent writes an output cache entry — a deep link to the deliverable — and calls the completion and context-feedback paths.

Memory and escalation

Each agent writes only to its own memory namespace and reads peer memory through context assembly. When a task needs governance the agent does not carry, it signals escalation with a recommended target. For parallelisable same-domain work, it signals a Shadow Clone with a sub-task breakdown.

On this page