What problem does it solve? When working in a Trellis-managed project, every implement or check phase must decide whether the main agent runs the work inline or dispatches a dedicated subagent. Without a router, the AI either asks repeatedly, guesses inconsistently, or loses the decision after context compaction. This Skill resolves the execution mode deterministically from session runtime state, a gitignored personal preference file, or auto-loop authorization, and only prompts the user when no valid decision exists. ## Core Features & Use Cases - Priority-based route resolution: Resolves implement/check routes in a fixed order (session runtime state, then .trellis/.route-prefs.tmp, then auto-loop authorization) via the bundled route_state.py helper, with compact JSON output for the agent. - Preference management: Lets users set, update, or clear personal default modes (inline vs subagent) per target, stored in a gitignored file that never enters commits. - Platform-aware subagent dispatch: Ships a platform-dispatch.json catalog covering 20+ AI platforms (Claude Code, Cursor, Codex, Gemini, etc.) that defines how to launch trellis-implement and audit-only trellis-check-all subagents, blocking silently unsafe fallbacks. - Use Case: During Phase 2.1 of a Trellis workflow, the main agent invokes this Skill; it recovers the session's prior route decision after a context compaction and outputs a structured route_decision telling the agent to dispatch trellis-implement as a subagent—without re-asking the user. ## Quick Start Ask the AI to route the current Trellis task's implement phase, for example: "Use trellis-route to decide whether this task's implementation should run inline or as a subagent."
