| name | minesweeper-autoplay |
| description | 运行、长测、诊断和优化 Windows 扫雷自动游玩流程。用于用户要求自动打开扫雷、自动玩、跑单局或多局测试、监控长测、分析失败截图、排查识图/几何/点击/策略问题、或继续迭代 `apps/minesweeper_autoplay/workflows/minesweeper_solver.py` 时。 |
Minesweeper Autoplay
Core Assets
- App package:
../../../apps/minesweeper_autoplay/
- Main solver:
../../../apps/minesweeper_autoplay/workflows/minesweeper_solver.py
- Runtime artifacts:
../../../runtime/apps/minesweeper_autoplay/test_artifacts/
- Knowledge closeout runs:
../../../runtime/knowledge_runs/
- Runner script:
../../../apps/minesweeper_autoplay/scripts/run-solver.ps1
- Formal project entry:
../../../skills/task_skills/minesweeper_autoplay/README.md
Standard Workflow
- Confirm the requested run mode.
single: 单局跑到自然胜负或单局结束即停
retry: 最多跑 N 局
until_success: 持续重试直到成功,或直到额外限制命中
- Start the solver with
apps/minesweeper_autoplay/scripts/run-solver.ps1 unless there is a good reason to call the Python entry directly.
- Automatic knowledge closeout is meaningful-only by default. Normal low-value short failures are skipped; successful, long-running, or boundary-signaling runs are closed out automatically.
- Monitor runtime artifacts instead of guessing what happened.
- Classify failures before editing code:
- recognition / geometry / click targeting
- strategy / pending rules / guessing quality
- run-control / retry-policy / stop-policy mismatch
- When changing the solver, patch
../../../apps/minesweeper_autoplay/workflows/minesweeper_solver.py, run py_compile, then rerun a targeted smoke test before longer tests.
- Keep the historical temporary solver path only for compatibility; new work belongs in the app package unless the user explicitly asks for platformization.
Run Commands
Use the wrapper script for normal operation:
powershell -ExecutionPolicy Bypass -File apps/minesweeper_autoplay/scripts/run-solver.ps1 -Mode single
Read references/run-modes.md for the supported parameter combinations.
Read references/usage.md for common user-facing invocation patterns.
Diagnostic Order
- Read
../../../runtime/apps/minesweeper_autoplay/test_artifacts/solver_stop_summary.txt when present.
- Inspect the latest
attempt_*_step_*_consensus_01.png and terminal failure screenshots.
- Check
cell_action_diagnostics.txt for no_effect, blocked, wrong_target, or visual-sync clues.
- Decide whether the issue is:
- board reading / classification
- click placement / geometry drift
- rule scheduling / safe moves not exhausted
- late-game guess quality
- run mode / stop condition mismatch
Read references/diagnostics.md and references/artifacts.md when diagnosing non-trivial failures.
Boundaries
- This skill formalizes the task family; it does not by itself promote code into
platform/src/omniauto/skills/.
- Formal platformization requires separate user approval.
- Keep human-readable catalog updates in
../../../skills/ aligned with runtime bundle updates in .agents/skills/.
References
references/run-modes.md
references/usage.md
references/diagnostics.md
references/artifacts.md