AgentSkillsCN

diff-runner

Oracle of Secrets 的回归测试工具。记录游戏过程场景，并在新版本中回放，以检测逻辑漏洞或同步异常。

SKILL.md
--- frontmatter
name: diff-runner
description: Regression testing tool for Oracle of Secrets. Records gameplay scenarios and replays them on new builds to detect logic bugs or desyncs.

Diff Runner

Scope

  • •Regression Testing: Verify that logic changes didn't break existing behavior.
  • •State Comparison: Compares critical game state (Link pos, Mode, Inventory) frame-by-frame (or action-by-action).

Core Capabilities

1. Record

Execute a scenario script and capture the "Golden Trace".

  • •record scenario.json trace.json

2. Verify

Replay a trace on the current ROM and assert state matches.

  • •verify trace.json

Workflow

  1. •Create Scenario: Write a JSON list of inputs: [{"input": "Right", "frames": 60}, ...].
  2. •Baseline: Run record on the stable ROM build.
  3. •Dev: Apply your changes/patches.
  4. •Test: Run verify on the new ROM build.
  5. •Result: "Pass" or "Divergence at Step 45".

Dependencies

  • •Tool: ~/src/hobby/yaze/scripts/ai/diff_runner.py.
  • •Mesen2: Running with socket server.

Example Prompts

  • •"Record a baseline trace for the 'walk_to_dungeon' scenario."
  • •"Verify the current build against the 'boss_fight' trace."

Troubleshooting

  • •Desyncs: Emulator RNG or uninitialized RAM can cause non-deterministic behavior. Ensure the scenario includes a hard reset or state load at the start.