vishal-kumar9032

Agent Context Guard

Name: Agent Context Guard
Author: vishal-kumar9032

Testing community

SOURCE https://github.com/vishal-kumar9032/agent-context-guard

Description

Codebase AI Agent Readiness & Context Hygiene Radar: audits steering docs (AGENTS.md, CLAUDE.md, .cursorrules), token waste/pollution, test determinism, and secret isolation. Offline, ~200ms, zero tokens.

Installation

This entry records only its repository, not the path inside it, so there is no exact command to give. Open the source below and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

README

agent-context-guard 🤖🛡️

**Codebase AI Agent Readiness & Context Hygiene Radar**

A fast, deterministic, 100% offline Rote Play that audits any repository to measure how effectively autonomous AI coding agents (Antigravity, Cursor, Claude Code, Copilot, Codex, Aider) can operate inside it.

Built for the **Rote Playoffs Hackathon 2026** by [@vishal-kumar9032](https://github.com/vishal-kumar9032).

⚡ The Problem

When developers open an AI coding agent inside a project, agents frequently hallucinate, degrade in reasoning quality, burn hundreds of thousands of unnecessary tokens, or leak credentials. Why?

No Steering Context: Missing AGENTS.md, CLAUDE.md, or .cursorrules forces agents to guess build, test, and style conventions.
Context Pollution: Giant unignored build folders, minified bundles (bundle.min.js), database dumps, or log files drown the LLM's context window in noise.
Missing Verification Scripts: Agents cannot run tests to verify their own code changes if there is no deterministic test command declared.
Secret Exposure in Prompts: Local .env files not excluded in .gitignore get ingested directly into remote model prompts.

🛡️ What `agent-context-guard` Checks

Pillar	Focus	What is Audited
🧭 Steering (25 pts)	Instructions	Presence and size of `AGENTS.md`, `CLAUDE.md`, `.cursorrules`, `.cursor/rules/*.mdc`. Flags bloated (>400 line) rulebooks that degrade model attention.
🧹 Token Hygiene (25 pts)	Context Noise	Scans for unignored `node_modules`, build directories (`dist`, `.next`, `target`), large binaries, and minified files. Estimates wasted tokens.
🧪 Verification (25 pts)	Determinism	Detects reproducible test, build, and lint commands (`npm test`, `pytest`, `cargo test`, `go test`, `make test`).
🔒 Safety (25 pts)	Secret Isolation	Detects unignored `.env` or credential files that could be exposed to LLM prompts.