菜单
Create or modify an LLM harness that subclasses BaseHarness. Use when implementing LLM context building, directive parsing, system prompts, or post-game analysis hooks for a game plugin. Triggers on "harness", "LLM context", "system prompt", "directive", "narrator", "advisor".
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Create or modify a game brain (decision engine) that subclasses BaseBrain. Use when implementing per-tick decision logic, adding roles, tuning thresholds, or writing the decide() method for a new game plugin. Triggers on "brain", "decide", "role", "threshold", "heuristic", "decision logic".
Configure or run the self-improving code evolution pipeline. Use when setting up evolution for a new game, customizing editable files, tuning the evolution prompt, running manual evolution, or managing the weekly fork/reset lifecycle. Triggers on "evolution", "evolve", "self-improving", "str_replace", "fork", "reset", "promote", "weekly cycle".
Create or modify the game policy (control loop) that subclasses BasePolicy. Use when implementing perceive() to parse game engine observations, execute() to convert commands to game actions, or wiring brain + harness together. Triggers on "policy", "perceive", "execute", "control loop", "game engine bridge".
Run or configure post-game analysis on completed game memory dumps. Use when setting up the analysis pipeline, customizing the analysis prompt, reviewing learnings files, or running manual analysis. Triggers on "analyze", "post-game", "learnings", "score", "review game", "what went wrong".
Create or modify game event triggers that subclass BaseEventDetector. Use when implementing event detection for LLM consultation, adding new trigger types, tuning priorities, or configuring debounce. Triggers on "trigger", "event detection", "consult", "LLM scheduling", "when to call LLM".
Create or update wiki knowledge pages for a game plugin or the framework. Use when writing strategy docs, skill guides, mechanics references, or updating the transferable knowledge base. Triggers on "wiki", "strategy doc", "skill doc", "mechanics doc", "game knowledge", "write documentation".
| name | cyborg-harness |
| description | Create or modify an LLM harness that subclasses BaseHarness. Use when implementing LLM context building, directive parsing, system prompts, or post-game analysis hooks for a game plugin. Triggers on "harness", "LLM context", "system prompt", "directive", "narrator", "advisor". |
The harness is a background daemon thread that monitors the game and consults the LLM when triggers fire. It builds context, calls the LLM, and parses the response into a Directive the brain can follow.
Subclass framework.base_harness.BaseHarness and implement four methods:
from framework.base_harness import BaseHarness
from framework.base_memory import GameMemory
from framework.types import Directive
class MyGameHarness(BaseHarness):
def parse_directive(self, data: dict, tick: int) -> Directive:
"""Parse LLM JSON response into a Directive."""
def build_context(self, memory: GameMemory, trigger: str,
operator_notes: list[str] | None = None) -> str:
"""Build the user message for the LLM consultation."""
def get_system_prompt(self, prior_learnings: str = "") -> str:
"""Return the LLM system prompt."""
def _run_post_game_background(self, dump, filepath, trigger):
"""Called at game end for analysis (optional)."""
The LLM responds with JSON. Validate and convert it:
def parse_directive(self, data: dict, tick: int) -> Directive:
role = data.get("role", "default")
if role not in VALID_ROLES:
role = "default"
return Directive(
role=role,
command=data.get("command"),
target=self._parse_coord(data.get("target")),
reasoning=data.get("reasoning", ""),
issued_tick=tick,
expires_tick=tick + 300,
)
Structure the context for scannability:
[TRIGGER: {trigger_name}]
[CURRENT STATE]
Position, HP, resources, score...
[RECENT EVENTS]
Last 10-15 episodic events
[STRATEGIC FACTS]
Active facts from strategic memory
[ACTIVE DIRECTIVE]
Current directive, age, status
[PERFORMANCE]
Rates and trends from perf windows
Tips:
Structure:
1. Identity: "You are a strategic advisor for a {game} agent"
2. Game rules: compact mechanics summary
3. Available actions: roles and commands the LLM can issue
4. Response format: strict JSON schema with examples
5. Cross-game learnings: injected via prior_learnings param
6. Wiki knowledge: loaded from game's wiki/ directory
Keep under 4000 tokens. Put the JSON format last (closest to generation).
harness = MyGameHarness(
provider=llm_provider,
analysis_provider=opus_provider, # optional, for post-game
event_detector=my_trigger_detector,
game_name="my_game",
game_id="abc123",
max_steps=2500,
seed=42,
runs_dir="runs/my_game",
)
framework/wiki/skills/llm-integration.md — consultation model, token budgetsframework/wiki/skills/memory-management.md — what to include from each tierframework/wiki/mechanics/memory-tiers.md — memory tier technical reference