Agent skill

Optimize Crafter Policy

by Linzwcs in Linzwcs/EvoPolicyGym

Improve an EvoPolicyGym Policy Program for the additive RGB Crafter survival-development Benchmark.

MITAuto-check passed

Install Optimize Crafter Policy

skills CLI
$ npx skills add Linzwcs/EvoPolicyGym --skill optimize-crafter-policy -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install Linzwcs/EvoPolicyGym optimize-crafter-policy --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/Linzwcs/EvoPolicyGym.git skills-src && mkdir -p .claude/skills && cp -r skills-src/environments/crafter/crafter/skills/optimize-crafter-policy .claude/skills/optimize-crafter-policy && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
optimize-crafter-policy
GitHub stars
175
Token cost
~1.7k tokens
SKILL.md length
800 words
Files
1
Skills in repo
5
Repo updated
First seen
Licence
MIT

At a glance

Improve an EvoPolicyGym Policy Program for the additive RGB Crafter survival-development Benchmark.

  • Works in 6 steps: Submit the packaged baseline on a small… → Inspect the complete compressed… → Eliminate invalid Actions, exceptions,… → …
  • SKILL.md covers Respect the ABI, Actions, Achievement progression and Build a verifiable…, plus 3 more sections
  • Instructions only: no scripts, shell commands, URLs or credentials in SKILL.md

What it does

Optimize Crafter Policy is an agent skill from Linzwcs/EvoPolicyGym. Improve an EvoPolicyGym Policy Program for the additive RGB Crafter survival-development Benchmark.

Its SKILL.md is about 1.7k tokens, which your agent loads only when the skill is triggered. It is a single SKILL.md file with no bundled scripts.

The repository describes itself as: EvoPolicyGym is infrastructure for evaluating coding agents and generating training experience through Autonomous Policy Evolution. The licence is MIT.

Example prompts

  • “/optimize-crafter-policy”

Workflow steps

6 steps, taken from the first numbered list in SKILL.md.

  1. Submit the packaged baseline on a small batch.
  2. Inspect the complete compressed trajectories, achievement success table,
  3. Eliminate invalid Actions, exceptions, and timeouts first.
  4. Add visual parsing for nearby terrain, creatures, facing direction, and
  5. Add a compact Episode plan for resource dependencies, survival, and
  6. Change one capability at a time and compare repeated batches because small

What it can do on your machine

Read from SKILL.md and the folder at commit a3d9669. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    No scripts in the folder and no shell commands in SKILL.md.

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Optimize Crafter Policy loads about 1.7k tokens when it runs. Until then it costs about 31 tokens; SKILL.md has 800 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~31
When it runs · the whole SKILL.md, loaded when a task matches
~1.7k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from Linzwcs/EvoPolicyGym at commit a3d9669, republished under its MIT licence (© Linzwcs). 800 words, ~1,698 tokens.

Download SKILL.mdSave it as .claude/skills/optimize-crafter-policy/SKILL.md (or your agent's skills folder).
name
optimize-crafter-policy
description
Improve an EvoPolicyGym Policy Program for the additive RGB Crafter survival-development Benchmark.

Optimize a Crafter Policy

Maximize additive long-horizon survival and development across deterministic but hidden procedural worlds. The Policy receives only a TensorValue containing a 64 x 64 x 3 uint8 RGB frame. It never receives the Environment seed, global semantic map, player position, achievement counters, reward, or privileged inventory data.

Respect the ABI

  • Export make_policy(context) from policy.py.
  • Return an object with act(observation).
  • Decode TensorValue.data as row-major RGB bytes with shape (64, 64, 3).
  • Return an exact integer from 0 through 16. Booleans, floats, containers, and out-of-range integers are invalid.
  • A fresh Policy instance is created for every Episode. Keep only Episode-local memory between calls to act().
  • Read only public static configuration from context.environment_parameters. Never infer or search for hidden seeds.

Actions

text
0  noop                 9  place_furnace
1  move_left           10  place_plant
2  move_right          11  make_wood_pickaxe
3  move_up             12  make_stone_pickaxe
4  move_down           13  make_iron_pickaxe
5  do                  14  make_wood_sword
6  sleep               15  make_stone_sword
7  place_stone         16  make_iron_sword
8  place_table

do collects or drinks from the facing tile and attacks adjacent creatures. Placement and crafting Actions work only when their public in-game prerequisites are satisfied.

Achievement progression

The 22 achievements cover collection, survival, combat, placement, and crafting. Build capabilities in a reusable dependency order:

text
wood -> table -> wood pickaxe -> stone
stone -> stone pickaxe + furnace
stone pickaxe -> coal + iron
wood + coal + iron + nearby table/furnace -> iron pickaxe
iron pickaxe -> diamond

Swords improve combat survival. Water, cows, plants, sleeping, and daylight management protect health, food, drink, and energy. The RGB frame includes the local world view and inventory display; build explicit visual parsing and Episode memory rather than hard-coded seed routes.

Build a verifiable resource-facility-craft state machine

Treat progression as confirmed state transitions, not as a timer that blindly cycles through recipe Actions. Keep an Episode-local state machine with at least these concepts:

  • Resource belief: visually estimate inventory counts, but distinguish an unconfirmed estimate from a resource whose collection was confirmed by the next RGB frame or HUD change.
  • Facility state: remember whether a table or furnace placement was attempted, visually confirmed, and still believed to be adjacent. Moving away must invalidate adjacency rather than leaving a permanent table_near=True or furnace_near=True flag.
  • Progress stage: advance only after observing evidence for the prior dependency. A useful order is survival stabilization, wood reserve, table, wood tool, stone reserve, stone tool, furnace, coal and iron, iron tool, and diamond.
  • Recovery transition: if expected evidence does not appear after an Action, retry from a legal facing tile, reacquire the missing resource, or rebuild the facility. Do not mark an achievement complete merely because its Action was returned.
  • Survival interrupt: drinking, food, sleep, combat, and escape can temporarily preempt crafting without erasing the current progression stage.

Make the state machine auditable through behavior. For each revision, inspect the achievement event order in the per-Episode trajectory.jsonl.gz Artifacts and use artifact-manifest.json to locate the lossless NPZ observation chunks. Verify dependencies such as collect_wood before place_table, and place_table before make_wood_pickaxe. Re-evaluate a promising unchanged Program across additional Episodes before trusting a small batch score.

Avoid policies whose main behavior is periodic do, periodic sleep, or a fixed recipe loop. Those Actions are legal but do not demonstrate that the resource and facility preconditions were perceived and maintained.

Show full SKILL.md (327 more words)Show less

Optimize the actual score

Each transition after which the player remains alive earns one survival point and up to 0.1 weakest-vital credit. The complete return is:

text
survival + weakest-vital quality
+ absolute first-unlock dependency progress
+ bounded repeated productivity

Every additional survived step remains valuable through the configured horizon. First-unlock dependency weights remain exponential, so later tools and materials are much more valuable than opening achievements. Repeated successful events add at most 25 points and have public caps and diminishing returns. A Policy failure returns -max_episode_steps for that Episode. Treat differences inside the published 95% sampling interval as inconclusive rather than as demonstrated improvement.

Audit inherited controller bias

Treat the packaged baseline as disposable scaffolding, not as a behavioral prior. Before extending it, identify control flow that ignores the current RGB observation and replace it when it creates a repeated spatial or Action motif. In particular, do not inherit:

  • fixed clockwise or counterclockwise direction cycles;
  • expanding-square, spiral, or rectangular patrol routes;
  • a fixed number of do Actions after every movement regardless of the facing tile;
  • periodic placement or crafting attempts without visible prerequisite evidence.

Exploration should react to visible walkability, resources, facilities, creatures, and recent Episode-local movement evidence. Inspect selected NPZ frames and the Action sequence after each revision. Repeated local loops or a dominant Action-frequency pattern are controller defects unless the current observation and state explicitly justify them.

Iterate safely

  1. Submit the packaged baseline on a small batch.
  2. Inspect the complete compressed trajectories, achievement success table, and selected frames from the complete lossless NPZ observations.
  3. Eliminate invalid Actions, exceptions, and timeouts first.
  4. Add visual parsing for nearby terrain, creatures, facing direction, and inventory icons.
  5. Add a compact Episode plan for resource dependencies, survival, and exploration.
  6. Change one capability at a time and compare repeated batches because small achievement samples are noisy.

Keep submitted Policy source inside program/. Store non-submitted diagnostic scripts, selected frames, and derived notes in the Agent-owned analysis/ directory. Do not access Host paths, runtime internals, or unavailable Crafter info fields.

© Linzwcs, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

Just SKILL.md in environments/crafter/crafter/skills/optimize-crafter-policy of Linzwcs/EvoPolicyGym.

Open the folder on GitHubat commit a3d9669

Compare with similar skills

Optimize Crafter Policy next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Optimize Crafter Policy compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Optimize Crafter Policy this skillLinzwcs/EvoPolicyGym175—~1.7kAutomated safety check: PassMIT
SQL Optimizationgithub/awesome-copilot40k2 repos~2.3kAutomated safety check: PassMIT
Agent Performance Optimizerruvnet/ruflo74k2 repos~3.6kAutomated safety check: PassMIT
Database Optimizerdavila7/claude-code-templates33k8 repos~2.5kAutomated safety check: PassMIT
Prompt Optimizeraffaan-m/ECC276k2 repos~2.4kAutomated safety check: PassMIT
Cost Optimizeruvnet/ruflo74k—~997Automated safety check: NotesMIT

Similar skills

  • SQL Optimization

    github/awesome-copilot

    Official

    Universal SQL performance optimization assistant for comprehensive query tuning, indexing strategies, and database performance analysis across all SQL databases (MySQL, PostgreSQL, SQL Server…

    40k GitHub starsUsed in 2 repos~2.3k tokens
    DatabasesAuto-check passed
  • Agent skill for performance-optimizer - invoke with $agent-performance-optimizer

    74k GitHub starsUsed in 2 repos~3.6k tokens
    Auto-check passed
  • Database Optimizer

    davila7/claude-code-templates

    Expert database optimizer specializing in modern performance tuning, query optimization, and scalable architectures.

    33k GitHub starsUsed in 8 repos~2.5k tokens
    DatabasesAuto-check passed
  • Prompt Optimizer

    affaan-m/ECC

    分析原始提示,识别意图和差距,匹配ECC组件(技能/命令/代理/钩子),并输出一个可直接粘贴的优化提示。仅提供咨询角色——绝不自行执行任务。触发时机:当用户说“优化提示”、“改进我的提示”、“如何编写提示”、“帮我优化这个指令”或明确要求提高提示质量时。中文等效表达同样触发:“优化prompt”、“改进prompt”、“怎么写prompt”、“帮我优化这个指令”。不触发时机:当用户希望直接执行任…

    276k GitHub starsUsed in 2 repos~2.4k tokens
    DevelopmentAuto-check passed
  • Cost Optimize

    ruvnet/ruflo

    Analyze token usage patterns and recommend cost optimizations with estimated savings

    74k GitHub stars~997 tokensUpdated today
    AI & LLM EngineeringAuto-check: notes
  • Caveman Optimization Evaluator

    JuliusBrussee/caveman

    Turns a Caveman report-only optimization observation into one minimal code change and a paired baseline evaluation, after the operator picks which to pursue.

    111k GitHub starsUsed in 1 repo~1.2k tokens
    AI & LLM EngineeringAuto-check passed

More from Linzwcs/EvoPolicyGym

  • Optimize Balatro Policy

    Linzwcs/EvoPolicyGym

    Build, modularize, improve, test, and select a reward-aligned EvoPolicyGym Bot system for the Jackdaw Balatro Benchmark.

    175 GitHub stars~4.8k tokensUpdated 21 days ago
    Auto-check passed
  • Evopolicygym

    Linzwcs/EvoPolicyGym

    Operate and extend EvoPolicyGym through its public SDK. An agent skill from Linzwcs/EvoPolicyGym.

    175 GitHub stars~890 tokensUpdated 21 days ago
    Auto-check passed
  • Improve an EvoPolicyGym Policy Program for Crafter local-symbolic-v1 observations.

    175 GitHub stars~805 tokensUpdated 21 days ago
    Auto-check passed
  • Optimize Nethack Policy

    Linzwcs/EvoPolicyGym

    Improve an EvoPolicyGym Policy Program for the deterministic NLE NetHackScore Benchmark.

    175 GitHub stars~1.9k tokensUpdated 21 days ago
    Auto-check passed

Questions about Optimize Crafter Policy

What does Optimize Crafter Policy do?

Improve an EvoPolicyGym Policy Program for the additive RGB Crafter survival-development Benchmark. Optimize Crafter Policy is an agent skill from Linzwcs/EvoPolicyGym. Improve an EvoPolicyGym Policy Program for the additive RGB Crafter survival-development Benchmark.

How do I install Optimize Crafter Policy in Claude Code?

Run `npx skills add Linzwcs/EvoPolicyGym --skill optimize-crafter-policy -a claude-code`. Or copy the skill folder (environments/crafter/crafter/skills/optimize-crafter-policy in Linzwcs/EvoPolicyGym) into .claude/skills/optimize-crafter-policy in your project. Claude Code loads it when a task matches its description.

How do I install Optimize Crafter Policy in Codex?

Run `npx skills add Linzwcs/EvoPolicyGym --skill optimize-crafter-policy -a codex`. Or copy the skill folder (environments/crafter/crafter/skills/optimize-crafter-policy in Linzwcs/EvoPolicyGym) into .agents/skills/optimize-crafter-policy in your project. Codex loads it when a task matches its description.

Can I use Optimize Crafter Policy in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add Linzwcs/EvoPolicyGym --skill optimize-crafter-policy -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/optimize-crafter-policy, .gemini/skills/optimize-crafter-policy, .github/skills/optimize-crafter-policy and .opencode/skills/optimize-crafter-policy in your project.

What does Optimize Crafter Policy need to run?

SKILL.md names no scripts, command-line tools or credentials: Optimize Crafter Policy is instructions for the agent only.

Does Optimize Crafter Policy access the network?

SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.

Is Optimize Crafter Policy safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.

What licence does Optimize Crafter Policy use?

Optimize Crafter Policy is published under the MIT licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Optimize Crafter Policy use?

About 1.7k tokens (SKILL.md is roughly 6.8k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.

What are the alternatives to Optimize Crafter Policy?

Skills that share tags, products or a category with Optimize Crafter Policy: SQL Optimization (github/awesome-copilot, 40k stars), Agent Performance Optimizer (ruvnet/ruflo, 74k stars), Database Optimizer (davila7/claude-code-templates, 33k stars) and Prompt Optimizer (affaan-m/ECC, 276k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Optimize Crafter Policy?

Linzwcs (a GitHub user) maintains it in Linzwcs/EvoPolicyGym, which has 175 GitHub stars. The repository holds 5 skills in this directory. The repository was last updated on September 18, 2026.

Source: Linzwcs/EvoPolicyGym on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.