Skip to content

Releases: DenisSergeevitch/game-sensitivity-coach

Game Sensitivity Coach v1.0.0

Choose a tag to compare

@DenisSergeevitch DenisSergeevitch released this 06 Sep 19:21

Game Sensitivity Coach 1.0.0

An agent skill that discovers mouse settings, transfers a familiar sensitivity between games, and reviews ordinary gameplay to recommend evidence-backed adjustments.

Install

Download game-sensitivity-coach.zip below and extract the entire folder into your agent's skill directory:

  • Codex: ~/.agents/skills/game-sensitivity-coach/
  • Claude Code: ~/.claude/skills/game-sensitivity-coach/

In Codex, mention $game-sensitivity-coach. In Claude Code, invoke /game-sensitivity-coach, then describe your game and what you want to improve. See the README for full setup and example prompts.

Included

  • Passive Windows gameplay recording with foreground gating and raw mouse input.
  • HTML evidence reports and exact-frame image exports for the host agent's visual review.
  • Settings discovery, explicit sensitivity conversion, and guarded configuration changes with backups and rollback.
  • Practical examples for THE FINALS, Battlefield 6, and Holdfast; synthetic fixtures and regression tests.
  • A reproducible skill ZIP, internal file/hash manifest, and separate SHA-256 checksum.

Requirements and scope

Python 3.8+ and an image-capable agent with local command/file access. FFmpeg is required for recording and frame export. Bundled live capture requires native Windows Python; offline helpers are portable. No third-party Python packages or separate AI API key are required.

The host agent performs visual analysis. The tools do not automate aiming or establish a universal best sensitivity. Live gameplay compatibility and proposed improvements require a fresh observation on the target computer.

Verify the download

Compare the SHA-256 digest of the ZIP with the attached game-sensitivity-coach.zip.sha256 file. The archive also includes RELEASE-MANIFEST.json with a hash for each packaged file.

Code and documentation: MIT License.