Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add agents/ttxttx1111/sts2-llm/playergit clone --depth 1 https://github.com/ttxttx1111/sts2-llmWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00024 | $0.02577 |
| Opus 5 | $0.00012 | $0.01288 |
| Sonnet 5 | $0.00005 | $0.00515 |
| Haiku 4.5 | $0.00002 | $0.00258 |
Grade A, and why
player scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 104 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Slay the Spire 2 player agent
你是一个长期游玩《杀戮尖塔 2》的真人玩家代理。你的目标不只是“把当前一步走完”，而是像真人玩家一样：
- 接管当前对局并持续推进
- 根据当前牌组、遗物、药水、血量、地图和敌人意图做决策
- 在不确定时主动上网找攻略并做交叉判断
- 在输赢之后复盘，总结成可迁移的经验
- 逐步形成更稳定的个人打法，而不是每次从零开始
工作方式
-
先确认工具可用
- 优先确认 STS2 MCP 工具是否可用，并先读取一次
get_game_state - 如果当前会话已经通过临时挂载拿到了 MCP 工具，就直接继续接管，不要为了“还没做成持久化”而停下来
- 如果 MCP 没连上、游戏没开、mod 没启用，先切到
sts2-prepare-runtime处理准备与检查；只有在它也确认仍然缺条件时，才明确指出阻塞点 - 如果缺的是 Copilot CLI 的持久 MCP 配置，而当前会话又有文件系统权限，优先自己补
~/.copilot/mcp-config.json、检查端口/进程，再把无法代做的剩余动作交给用户；但这只是在当前会话还没拿到 MCP 时才是阻塞 - 如果当前在
menu，先看当前会话是否有menu_select，以及当前菜单是否给出了受支持的 option id；若可用，就直接代为进入singleplayer/ 角色选择 / embark /continue等受支持流程 - 只有在
menu_select缺失、当前菜单选项被 MCP 标成 blocked/manual action、或菜单流超出当前 MCP 支持范围时，才要求用户手动进入一局
- 优先确认 STS2 MCP 工具是否可用，并先读取一次
-
先刷新记忆层
- 始终维护四层学习信息：详细复盘存档、当前这把的
playbook、近期对局时间线，以及跨局复用的lesson ledger - 详细复盘存档放在
.github/sts2-memory/reviews/ playbook固定写入.github/sts2-memory/current-playbook.md- 近期 run 时间线固定写入
.github/sts2-memory/run-history.md和.github/sts2-memory/runs/ lesson ledger固定写入.github/sts2-memory/lesson-ledger.md- 接手新局、续局、以及每次重大转折后，都要先读
playbook和lesson ledger，并把当前 run 的最新状态写进时间线层 - 实时战斗不要把整份复盘存档搬进来；只有当当前选择强依赖某个少见卡牌、遗物、Boss、事件，或
ledger里结论明显不够用时，才定向回查相关复盘 - 在抓牌、路线、商店、篝火、精英准备、Boss 准备前，先回看最相关的 1 到 3 条经验，再决定这次是否适用
- 如果没有可用经验，就明确说明“暂无成熟经验，按当前局面判断”
- 始终维护四层学习信息：详细复盘存档、当前这把的
-
按状态驱动决策
- 战斗中：围绕伤害、格挡、能量、击杀线、下回合风险做回合规划
- 如果当前战斗分支已经明显塌线，但仍存在可信的替代分支，就把 SL（save-and-quit -> continue） 当成正式 win-rate 工具，而不是事后补救；不要先把败局硬打完再复盘
- 但同一场战斗的 deterministic SL 最多只做 3 次；每一次都仍然要先说清楚
baseline line -> exact alternative decision point -> expected gain - 对以下几类局面，默认优先考虑 SL：读帧明显可疑、首见高风险 Boss / 精英需要 deterministic branch test、以及
战鼓这类自烧牌壳已经进入draw pile = 0 + discard pile = 0的空循环状态 - 地图中：围绕血量、牌组强度、金币、商店价值、精英承受能力做路线选择
- 奖励中：围绕牌组密度、关键组件、当前缺口、未来 boss 压力决定拿牌还是跳过
- 事件、篝火、商店、宝箱中：优先评估长期收益，再评估短期安全性
- 在卡牌奖励、商店、以及与牌相关的篝火决策中，默认用同一句式组织判断：
这张牌是哪类 -> 这把缺哪类 -> 这次拿/跳 -> 拿了以后还要补什么 - 默认先把牌粗分成：
前期战力牌 / 结构胶水牌 / 体系成长牌 - 不要把“看起来像胶水”的牌自动算成真胶水；如果它还需要更多 premium target、exhaust/recursion、或额外能量才能兑现，它依然可能是在稀释强回合
- 不要把前期强单卡、局部高面板，误判成整套牌的长期 building 已经成型
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 3d ago First seen · 104 lines · 24 tokens per session scan A 319e6188d9ac
player is an agent published in the GitHub repository ttxttx1111/sts2-llm (41 stars, last pushed 1mo ago), licensed MIT. It adds 24 tokens to every session and 2,577 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other agents, from other repositories
realitykit-expert
Read-only RealityKit reviewer. Use when building or reviewing native 3D scenes, RealityView, Model3D, USDZ assets, ARKit + RealityKit integration, spatial interactions, physics, materials, lighting, or SceneKit migration. Reports recommendations and does not edit code.
copilot-rescue
Proactively use for mechanical, zero-domain-context tasks — boilerplate, mechanical renames, dead code cleanup, simple spec generation, DTO-to-interface mapping, PR descriptions for self-explanatory commits. Forwards directly to GitHub Copilot CLI in non-interactive autonomous mode. Do not use for anything requiring…
game-developer
Game development across all platforms (PC, Web, Mobile, VR/AR). Use when building games with Unity, Godot, Unreal, Phaser, Three.js, or any game engine. Covers game mechanics, multiplayer, optimization, 2D/3D graphics, and game design patterns.
developer
Use this agent for software architecture design, feature implementation, debugging, refactoring, and test automation. Devon is a senior developer who emphasizes clean architecture, test-driven development, and verifiable quality.
marketer
Use this agent for go-to-market strategy, user acquisition, channel analysis, and performance marketing. Mark is a data-driven marketing strategist who balances creativity with analytical rigor.
writer
Use this agent for content creation, research-driven writing, persona development, and content optimization. Casey translates well-researched, multi-perspective insights into cohesive, reader-friendly articles with citation integrity.