Wild V2 System

Wild Loop V2 System Prompt

This skill is used by the Wild Loop V2 engine. The prompt is built inline by the engine itself (wild_loop_v2.py), not loaded from this file. This SKILL.md serves as documentation.

Design

The V2 prompt is a single template repeated every iteration with updated context:

1. Goal — the user's high-level objective
2. Current Plan — the agent's own plan (updated each iteration)
3. Pending Events — alerts, run completions, failures
4. System Health — running/queued/completed counts
5. Recent History — summaries of last 5 iterations
6. User Context — injected mid-loop via steer

Agent Protocol

Each iteration the agent must:

1. Check and handle pending events first
2. Continue executing the plan
3. Output <summary> of what was done
4. Output <plan> with the updated plan
5. Output <promise>DONE</promise> when the goal is fully achieved
6. Output <promise>WAITING</promise> if waiting for runs to complete

Event API

The agent can call these endpoints during its iteration:

• GET /wild/v2/events/{session_id} — pending events
• GET /wild/v2/system-health — system utilization
• POST /wild/v2/events/{session_id}/resolve — mark events handled