mobilegym: Skill for Claude Code

.claude/skills/writing-bench-task-judge/SKILL.md

writing-bench-task-judge is a skill for Claude Code from Purewhiter/mobilegym. It costs 68 tokens per session (4,584 once invoked), scanned A, original, Apache-2.0.

Guidance for writing judges that decide whether an agent completed a benchmark task. It focuses on checking the actual required changes rather than relying on fragile words or incidental details.

In plain words
What is it for?
It helps add or review task checks, classify tasks as creating or deleting data, compare the initial and final state, and use the benchmark's task and judge specifications correctly.
Why use it?
It reduces false passes for incomplete work and false failures for valid solutions.

Skill for Claude Code

Written for Claude Code: installed under .claude/.

This is Purewhiter/mobilegym's own configuration. It tells Claude Code how to work on mobilegym itself, so it is not a mod to install elsewhere. Copy it as a starting point and replace the rules that are about this project. Everything mobilegym configures →

Reuse

Borrowing it

Nothing to install: this file belongs to Purewhiter/mobilegym. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.

Copy the file
curl -O https://raw.githubusercontent.com/Purewhiter/mobilegym/main/.claude/skills/writing-bench-task-judge/SKILL.md
Clone the repo
git clone --depth 1 https://github.com/Purewhiter/mobilegym

Made for: Claude Code.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for writing-bench-task-judge

README.md
[![agentmods](https://agentmods.dev/badge/skills/purewhiter/mobilegym/writing-bench-task-judge/github.svg)](https://agentmods.dev/skills/purewhiter/mobilegym/writing-bench-task-judge)
Your own site
<a href="https://agentmods.dev/skills/purewhiter/mobilegym/writing-bench-task-judge"><img src="https://agentmods.dev/badge/skills/purewhiter/mobilegym/writing-bench-task-judge/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for writing-bench-task-judge

Your own site · 80×15
<a href="https://agentmods.dev/skills/purewhiter/mobilegym/writing-bench-task-judge"><img src="https://agentmods.dev/badge/skills/purewhiter/mobilegym/writing-bench-task-judge.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 68 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 4,584 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe. Third-party audits
  • NVIDIA SkillSpector pass 7 Sept 2026
How audits are shown
Origin original No closer match found in the catalogue.
Token cost

What 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.

ModelPer sessionOnce invoked
Fable 5.1 $0.00068 $0.04584
Opus 5 $0.00034 $0.02292
Sonnet 5 $0.00014 $0.00917
Haiku 4.5 $0.00007 $0.00458

Measured 12d ago against content hash 46d72eb770b4, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-11, from the pricing page.

Security

Grade A, and why

writing-bench-task-judge 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 12d 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.

.claude/skills/writing-bench-task-judge/SKILL.md · 255 lines

How it starts

The opening of the file, as written. The whole thing — 255 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Writing bench_env Task Judges

Overview

A judge decides whether the Agent completed the task. Two failure modes dominate:

  • Soundness hole — an obviously wrong path is judged passed=True (keyword hit, unstable trace field).
  • Completeness hole — a reasonable completion path is judged passed=False (bound to specific wording or UI step).

Neither is caught by type checks — both are caught by applying the CRUD model rigorously.

Authoritative refs:

  • bench_env/docs/task/TASK_AUTHORING_GUIDE.md — task design + CRUD judge model (read §2 in full)
  • bench_env/docs/task/TASK_CODE_SPEC.md — code rules (file responsibilities, naming, defensive-coding ban, time APIs, CriteriaTask)
  • bench_env/docs/REFERENCE.mdJudgeInput / JudgeResult field & expected_changes path syntax lookup
  • bench_env/docs/task/GROUNDED_MODE.md — only if task uses answer_fields / grounded eval

Step 1 — Classify the task as CRUD

Before writing any check, name the operation out loud:

Operation Triggering verbs Required check strategy
Create (增) 添加/创建/发送/收藏/点赞/新建 diff: init vs current, match in new items
Delete (删) 删除/移除/取消/取关/下架 diff: target id in init_ids - curr_ids
Modify (改) 修改/切换/改名/设置为/拨动 lookup in current; use init to resolve identity
Query (查) 告诉我/是多少/哪个/什么时候 read from init, compare to input.answer

Each CRUD type has ONE correct strategy — not a choice. Using lookup-on-current for Delete/Create creates false positives (sampler bug → passes as Agent action).

Step 2 — Survey the App module for reusable abstractions

Before writing any judge, Read app.py end-to-end with the Read tool (split into sections only if the file exceeds the tool's limit). Scan every def line plus its docstring / leading comment, plus any module-level helper functions above the class. Do not substitute grep for this. Grep silently misses:

  • module-level helpers (_note_text, _pick_keywords_from_note) that sit above the class
  • @staticmethod / @property / @classmethod methods that your prefix guess won't anticipate
  • helpers whose names don't follow the shape you grepped for (latest_note_by_title is a find-by-name primitive but doesn't start with find_; visible_notes is a filtered view but doesn't start with filtered_)
  • _xxx private methods that contain logic you could extract

Read the full file on GitHub · 255 lines

Changes

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.

  1. 12d ago First seen · 255 lines · 68 tokens per session scan A 46d72eb770b4

Subscribe to this mod's changes

writing-bench-task-judge is a skill published in the GitHub repository Purewhiter/mobilegym (787 stars, last pushed 14d ago), licensed Apache-2.0. It adds 68 tokens to every session and 4,584 once invoked, about $0.0003 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.

Related

Other skills, from other repositories

meter-benchmark

Measure per-stage success probability before iterating on CocoFlow templates. Supports $meter benchmark --flow .

Snowflake-Labs/cocoplus · 27 tokens

belt

Operate the belt CLI to evaluate headless coding agents (Claude Code, Cursor, Codex, Gemini, and others) end to end. Use when the user asks to write or run eval scenarios, compare agents, score outputs with rules or LLM judges, register a new agent adapter, interpret reports or benchmark cards, or set up evals in CI.…

jfrog/agent-belt · 116 tokens

foundry-hosted-agent-validation

Step-by-step process for validating a Python Foundry hosted agent sample (under python/samples/04-hosting/foundry-hosted-agents/) end to end — running it locally (native runtime and azd ai agent run) and after deploying it to an Azure AI Foundry project with azd. Use this when asked to validate a hosted agent sample.

microsoft/agent-framework · 82 tokens

build-and-test

How to build and test .NET projects in the Agent Framework repository. Use this when verifying or testing changes.

microsoft/agent-framework · 26 tokens

skill-release-gate

Evaluate an Agent Skill bundle for structural integrity, trigger quality, artifact improvement, script correctness, safety, installed-tree integrity, and target-host portability before release.

rohitg00/ai-engineering-from-scratch · 36 tokens

workflow-patterns

Use this skill when implementing tasks according to Conductor's TDD workflow, handling phase checkpoints, managing git commits for tasks, or understanding the verification protocol.

wshobson/agents · 35 tokens