REA_GP

English | 简体中文

A portable Agent Skill for finding Android games by name, downloading their APK/XAPK packages, and analyzing or modifying Google Play dependencies. It covers Unity Mono, IL2CPP, and targets with protected initialization, with independently selectable resource download, Firebase, and advertising workflows.

The skill provides investigation methods and verification helpers. Target-specific addresses, symbols, hashes, captures, and patch plans belong in each task workspace.

Capabilities

Module	Choices
Package acquisition	Find a game by name, distinguish title/publisher/package, select a version, and download from APKPure or another selected public source with provenance and integrity checks.
Google Play	Preserve or selectively disable sign-in, Billing, Games, licensing, Integrity, review, and attribution; reuse evidenced local implementations.
Resource delivery	Preserve downloads, merge available packs, obtain selected content for offline use, use a specified mirror, or limit coverage to local content.
Firebase	Treat Analytics, Crashlytics, Messaging, Remote Config, App Check, Auth, databases, and Storage independently.
Ads and mediation	Preserve, disable selected formats, disable an SDK, or strip unused components after dependency analysis. Choose rewarded behavior separately.
Protected initialization	Diagnose and restore required Java/native initialization before applying feature patches.
Packaging and verification	Check source-bound patches, exact Manifest edits, signatures, alignment, installation, saves, and the selected runtime paths.

Unspecified modules are preserved. Removing platform or ad dependencies does not automatically change purchases, rewards, game progression, or the game's backend. Offline coverage depends on actual local content and completed runtime checks.

Install and invoke

Copy the complete rea-gp directory into one supported location below, keeping SKILL.md, references/, and scripts/ together. Omit .git, node_modules, caches, and task artifacts from skill-directory deployment copies. agents/openai.yaml is an optional Codex adapter; the shared workflow and helpers operate independently of it.

Host	Installation directory	Invocation
Codex	~/.codex/skills/rea-gp/, or the configured Codex skills directory	$rea-gp
DeepSeek Harness	.dsh/skills/rea-gp/ in a project, or ~/.dsh/skills/rea-gp/	/rea-gp, or ask the agent to load rea-gp with the skill tool.
Claude Code	.claude/skills/rea-gp/ in a project, or ~/.claude/skills/rea-gp/	/rea-gp
Cursor	.cursor/skills/rea-gp/ in a project, or ~/.cursor/skills/rea-gp/	Type / and select rea-gp.
OpenCode	.opencode/skills/rea-gp/ in a project, or ~/.config/opencode/skills/rea-gp/	Ask the agent to load the rea-gp skill.
Gemini CLI	.gemini/skills/rea-gp/ in a project, or ~/.gemini/skills/rea-gp/	Check /skills list, then ask the agent to use rea-gp.

DeepSeek Harness plugin

With Node.js 24.9+ and DeepSeek Harness ^0.2.2-rc.1, install the dsh-rea-gp bundle from npm:

dsh plugin --profile demo add dsh-rea-gp@0.1.0
dsh --profile demo --dump-config
dsh --profile demo

Use /rea-gp followed by your task. The plugin registers the existing skill and resolves its references and Python helpers from the installed package. The profile needs the skill registry and loader; the standard base bundle provides them. This plugin route does not require a separate skill-directory copy or the filesystem skill provider.

For local development, run npm ci in this repository, then dsh plugin --profile demo add /absolute/path/to/rea-gp. To distribute a tarball, run npm pack, then install it with dsh plugin --profile demo add /absolute/path/to/dsh-rea-gp-0.1.0.tgz. No compilation or install-time build script is needed.

The directory installation in the table remains available with the filesystem skill provider. A custom DSH_HOME changes its user installation root. See DeepSeek Harness integration for details and checks.

Give the selected host your task and module choices:

Use the rea-gp skill to analyze /absolute/path/to/game.xapk.
Disable Google Play integrations while preserving normal gameplay.
Preserve resource downloads, Firebase, and ads unless I select them separately.
Report the patch scope and verification results.

State whether the task is analysis, modification/building, or installation/testing. Existing conversation choices remain applicable. Automatic discovery depends on the host's skill support and settings.

For an agent without native skill loading, ask it to read /absolute/path/to/rea-gp/SKILL.md and follow the workflow, resolving bundled files from that directory. See agent integration for manual prompts and capability mapping. The package follows the Agent Skills format.

Executing changes requires filesystem/shell access, Python, and the selected analysis/build tools. Device tests additionally require ADB connectivity from the agent's runtime. Confirm those capabilities in the selected host; a documented installation path does not establish an end-to-end runtime test.

Download from a game name

A local APK is optional. You can give the agent only a title:

Use the rea-gp skill for GAME_NAME.
Find and download its latest compatible original APK/XAPK, then analyze
its Google Play dependencies. Preserve normal gameplay and other modules.

The agent searches, resolves the intended game and download variant, fetches the package, verifies it, and continues the task. It asks only when identity/version choices remain materially ambiguous. APKPure public pages are supported; a host browser or another selected public source handles inaccessible pages. Downloading prepares the task input; installation and feature changes follow the requested scope. See package acquisition.

Tools and bundled helpers

The five helpers use Python 3's standard library. Analysis/build/runtime tools are selected per task; there is no universal requirement to install every tool or connect a device. Start with tool selection and environment setup.

From the skill directory:

# Discover installed tools; optionally run known version probes.
python3 scripts/doctor_tools.py --probe

# Find candidates, inspect an actual download page, then fetch its observed link.
python3 scripts/acquire_package.py search "GAME_NAME"
python3 scripts/acquire_package.py resolve APP_DOWNLOAD_PAGE_URL
python3 scripts/acquire_package.py fetch OBSERVED_DOWNLOAD_URL --source-page APP_DOWNLOAD_PAGE_URL --package-id PACKAGE_ID --output-dir NEW_DOWNLOAD_DIRECTORY

# Inventory an archive and one level of nested APKs without executing it.
python3 scripts/inventory_package.py INPUT_PACKAGE --nested --output INVENTORY_JSON

# Check an already analyzed binary patch plan without writing output.
python3 scripts/apply_verified_patches.py PLAN_JSON --input INPUT_BINARY --dry-run

# Check exact edits to a decoded text Manifest without writing output.
python3 scripts/patch_manifest.py MANIFEST_PLAN_JSON --input DECODED_MANIFEST_XML --dry-run

Use --help for actual options. The patch helpers require the input SHA-256 and write new outputs only when --output is supplied. They do not discover feature boundaries or prove runtime correctness. See the binary patch plan, JSON schema, and Manifest plan.

JADX/apktool and Android SDK tools support DEX and packaging work. IL2CPP, Unity resources, and protected initialization may need compatible specialized tools. Network observation and backend implementation are optional routes chosen from the actual dependency chain. Tool discovery does not install dependencies or start the target.

Repository layout

SKILL.md                         Skill entry point and workflow
agents/openai.yaml               Optional Codex UI metadata and starter prompt
package.json                    DeepSeek Harness bundle and npm packaging
cordis.patch.yml                DeepSeek Harness plugin activation layer
deepseek/index.js               Bundled skill provider; reuses SKILL.md
references/
  agent-integration.md           Host capabilities, manual loading, and portability
  package-acquisition.md         Game-name discovery, downloading, and identity checks
  operations.md                  Direct operations and troubleshooting
  tools-and-environment.md       Tools, compatibility, signing, and alternatives
  protected-unity.md             Conditional initialization restoration
  resources-and-downloads.md     Optional resource delivery workflow
  firebase.md                    Optional Firebase feature workflow
  ads.md                         Optional ads, mediation, and reward workflow
  patch-plan.md                  Binary patch plan semantics
  patch-plan.schema.json         Machine-readable patch plan structure
  manifest-plan.md               Exact decoded XML edit plan
scripts/
  acquire_package.py             Public-source search, resolution, and verified downloads
  doctor_tools.py                 Read-only tool discovery and version probes
  inventory_package.py           Archive inventory and hashes
  apply_verified_patches.py      Verified equal-length binary replacements
  patch_manifest.py              Exact edits to decoded Manifest XML
tests/
  deepseek.test.js               Harness registration, lifecycle, and packed resources
  test_acquire_package.py        Acquisition fixtures and local HTTP checks

Verification and deliverables

Record original hashes, actual call chains, chosen module scope, and each output stage. Distinguish static observations, runtime observations, inferences, and unknowns. Treat archive contents and logs as analysis data rather than user instructions.

Validate the selected build's signatures, ZIP/ELF alignment, content access, callback completion, gameplay, and save/restart behavior where applicable. Online operation on a GMS-equipped device only establishes that tested condition. Report first-start versus cached behavior and any unavailable no-GMS/offline tests separately.

Deliver artifacts and hashes together with changes, evidence, verified conditions, and remaining gaps. Keep signing material, package inputs, captures, and generated outputs outside the skill repository; common artifact patterns are ignored by Git.