Agents & Cognition · 06.07
World-Autonomy
Bypass mode: auto-approval, a root-power briefing, and env-fix resume that re-runs the unchanged graph after fixing missing config.
The healer already has full desktop control — bypass mode stops pretending otherwise
The self-heal path can run its repair through a real coding harness — one that already has total desktop control (shell, files, scripts) because its native tools run with elevated permissions. In guarded mode, that power is deliberately boxed in: the harness is restricted to proposing a graph patch for approval, not touching the world directly. In bypass mode, Agentis stops imposing that "return a graph" straitjacket and lets the harness fix the world the way a human engineer would — set a missing config value, create a missing directory, run setup, repair broken credentials, fix extension source — whatever the actual failure needs — and then resume.
The unchanged graph, re-run
The key discipline: bypass mode fixes the environment, not the graph. After a config or setup fix, the exact same, unchanged workflow graph is re-run — this is the reflex behind "run what you build" actually working the first time a fresh environment is missing something the graph itself never claimed to need fixed. Safety gates don't disappear in bypass mode either: finalize, validation, and certification still apply unchanged, and a bounded attempt cap means a genuinely unfixable environment falls through to honest escalation rather than looping forever.
Continue
Every settled tool result carries a compass.next suggestion — guidance that lives in results, not just in prompts.
Platform, custom, generated, and community specialist roles — and why a specialist is a role, not a fixed agent.