AgentSkillsCN

run

在Play模式下，操控AI智能体在当前场景中运行——测试导航、战斗、拾取物、陷阱，并报告诸如无法到达的区域、卡住的状态，或触发器失效等问题。

SKILL.md
--- frontmatter
name: run
description: Drive an AI agent through the current scene in Play mode — test navigation, combat, loot, traps, and report issues like unreachable areas, stuck states, or broken triggers.
allowed-tools: Bash, Read, Write, Edit, Glob, Grep
argument-hint: <test focus, e.g. "navigation test" or "full dungeon run">

Run an AI playtest session: $ARGUMENTS

Process

  1. •Analyze the scene — get hierarchy, find player spawn, exits, enemies, loot, traps, triggers
  2. •Enter Play mode via MCP
  3. •Execute test sequence:
    • •Navigate between key points (spawn → exits, spawn → loot, spawn → enemies)
    • •Screenshot periodically for visual analysis
    • •Monitor console for errors and warnings
    • •Track time spent in each area
  4. •Test categories:
    • •Navigation: Can the player reach all exits? Any stuck points?
    • •Combat: Do enemies spawn? Can they be damaged? Do they die properly?
    • •Loot: Do items drop? Can they be collected? Do they grant effects?
    • •Traps: Do triggers fire? Do damage zones work? Can they be avoided?
    • •UI: Do health bars update? Does inventory reflect changes?
  5. •Exit Play mode
  6. •Generate report:
    • •Issues found (with screenshots)
    • •Areas tested vs. untested
    • •Performance observations
    • •Suggested fixes