Vol. 01 — 2026
🧠

Developer Tooling

Agent Context Architect

Compile CLAUDE.md, AGENTS.md, and Cursor rules into one budgeted context system.

★ 4.8 71 developers Sub-300 token kernel

Your AI coding agent is not ignoring your instructions — it is drowning in contradictory rules. Agent Context Architect audits your messy repository instructions, detects syntax and style collisions across five families, and compiles them into a sub-300 token core kernel.

⚡ Direct Answer & Architecture Blueprint

A compiler and linter that detects style and command collisions across repository rules and compiles them into a budgeted sub-300 token core kernel.

Target Audience

Full-stack developers, engineering leads, and DevOps teams using Cursor, Claude Code, Codex, and Antigravity on large codebases.

Core Mechanism

Deterministic collision detection across package managers, linters, and quote styles, generating pointer-only client adapters.

Verified Benchmark

Reduces context injection overhead from 8,000+ tokens to <300 tokens, saving up to 70% in repetitive prompt API costs.

Why this architecture wins

Without this skill (Manual & Flawed) With this skill (Deterministic & Gated)
✕ AI coding agents hallucinating or ignoring instructions due to contradictory rule files. ✓ Single compiled sub-300 token kernel with pointer adapters for Claude, Cursor, and Codex.
✕ Rules drifting out of sync across multiple developer machines. ✓ Strict CI drift gate blocking PRs that introduce unbudgeted instruction bloat.
✕ Spending thousands of recurring tokens sending redundant system prompts. ✓ Sub-300 token core kernel slashing token burn while improving compliance.

What's inside

  • ✓skills/context-architect/SKILL.md (Audit, adjudicate, compile pipeline)
  • ✓skills/context-drift-guard/SKILL.md (CI maintenance & drift triage)
  • ✓audit-context.mjs (Deterministic context audit with JSON/Markdown reports)
  • ✓compile-context.mjs (Budgeted kernel generator + multi-client adapters)
  • ✓verify-context.mjs (Strict CI gate blocking unbudgeted context bloat)
  • ✓Messy Next.js reference repo fixture + expected compiled outputs

Built with

  • Node.js 18+
  • Markdown
  • Claude Code
  • Cursor Rules
  • CI Gates

How to use this skill (Step-by-step)

Run offline in your local workspace, orchestrate through any agent (Claude Code, Cursor, Antigravity, Cline), or plug directly into CI/CD pipelines.

1

Unzip Compiler into Repository

Place the tool scripts into your project root or developer utilities folder.

unzip agent-context-architect-1.0.0.zip -d ./context-tools
2

Audit Existing Repo Instructions

Scan CLAUDE.md, .cursorrules, and instructions for style collisions.

node ./context-tools/scripts/audit-context.mjs .
3

Compile into Sub-300 Token Kernel

Compile adjudicated rules and generate pointer adapters for all tools.

node ./context-tools/scripts/compile-context.mjs .
4

Add Drift Check to Git CI Pipeline

Ensure developers do not inadvertently bloat prompt kernels in pull requests.

node ./context-tools/scripts/verify-context.mjs --max-tokens=300

Included with every purchase

  • Complete compiler and auditing scripts with zero dependencies
  • Commercial license for unlimited personal and client repositories
  • Pre-configured adapters for Claude Code, Cursor, Codex, and Antigravity
  • Detailed step-by-step tutorial guide with 5 prompt recipes
  • 12 months of free minor updates

Questions

Will this overwrite my existing rule files?

No. The auditor is strictly read-only, and compilation creates timestamped backups of all existing adapter files before making any modifications.

Does it send my repository data over the internet?

No. All parsing, adjudication, and compilation runs 100% locally on your machine with zero network calls or telemetry.

Which languages and frameworks are supported?

The context auditor works on any codebase. Automated command extraction supports JavaScript/TypeScript package.json, with manual kernel overrides for Python, Go, Rust, and PHP.