ContextSect

Spend less. Ship more.

Agent-agnostic token optimization. One framework, every AI coding client. Evidence-based rules that save 45–60% tokens without reducing quality.

Get Started GitHub

The Problem

AI coding agents waste 40–80% of tokenson context they don't need, output nobody reads, and implementations that go in the wrong direction.

5k–50k tokens

Wrong directionVague prompt → agent guesses → wrong implementation → redo

40–80% input

Context pollutionReading full files when only one function matters

40–65% output

Output bloatFiller, restatements, full-file rewrites, narration

10x–100x normal

Runaway loopsStuck tool calls repeating without progress

30–50% waste

Session accumulationOld context never pruned, resent every turn

Key insight: Output tokens cost 5× input tokens ($15/M vs $3/M). A single unnecessary full-file rewrite wastes more money than reading 4,000 tokens of input.

Two-Pillar Architecture

ContextSect attacks the problem from both sides — reducing what goes IN and compressing what comes OUT.

PILLAR 1

Input Optimization

Prevents unnecessary context expansion BEFORE work begins.

→Alignment gate — catch vague prompts early
→Search-first — targeted reads only
→Progressive loading — tiers, not dumps
→Loop detection — halt stuck patterns

-40–55% input tokens

PILLAR 2

Output Optimization

Minimizes generated tokens AFTER reasoning completes.

→Zero filler — no pleasantries or narration
→Diff-only — SEARCH/REPLACE, never full files
→No restatement — start with the answer
→Explain only when asked

-50–65% output tokens

45–60%

total cost reduction (weighted by 5× output multiplier)

10 Agents. One Framework.

Write rules once → auto-adapted to each agent's native format on install.

Kiro

.kiro/steering/*.md + hooks

Claude Code

~/.claude/CLAUDE.md

Cursor

.cursor/rules/*.mdc

Windsurf

.windsurf/rules/*.md

Cline

.clinerules/*.md

OpenCode

~/opencode.md

Aider

.aider.conventions.md

RooCode

.roo/rules/*.md

GitHub Copilot

copilot-instructions.md

OpenAI Codex

~/AGENTS.md

Auto-detection: The installer scans your system for installed agents and configures each one — no manual config file editing needed.

8 Modular Rules

Each rule works independently. Enable what you need, disable what you don't.

Rule

Purpose

Input ↓

Output ↓

On?

output-contract

Zero filler, diff-only, no narration

—

-40–65%

✅

diff-only

SEARCH/REPLACE format, never full files

—

-60–90%

✅

search-first

Targeted search before file reads

-40–80%

-20–30%

✅

loop-breaker

Halt stuck execution patterns

∞ prevent

✅

alignment-gate

Clarify before complex tasks

-50–70%

⚡

plan-before-act

Plan multi-file changes

+5%

-50–70%

⚡

investigation-mode

Evidence-first debugging

-40–60%

-50%

⚡

context-hygiene

Progressive loading, session awareness

-30–50%

indirect

⚡

✅ Always on (zero risk)⚡ Task-specific (auto-triggered)

Configuration Profiles

Choose your intensity level. The installer configures everything automatically.

Conservative

Input ↓15–25%

Output ↓20–30%

RiskZero

Learning, prototyping

recommended

Balanced

Input ↓40–55%

Output ↓50–65%

RiskLow

Daily development

Aggressive

Input ↓60–75%

Output ↓70–85%

RiskMedium

Familiar codebases

Ultra-Aggressive

Input ↓80–90%

Output ↓85–95%

RiskHigh

Automated agents

$ contextsect profile balanced

Install

One command. Auto-detects your agents. Installs the CLI globally.

terminal

$ curl -sL https://contextsect.vercel.app/install.sh | bash

  ╭──────────────────────────────────────────────╮
  │   ContextSect — Token Optimization           │
  │   Agent-Agnostic • Evidence-Based • Modular  │
  ╰──────────────────────────────────────────────╯

  ✓ git 2.53.0
  ↓ Cloning ContextSect...
  ✓ Source ready at ~/.contextsect
  ✓ CLI installed: /usr/local/bin/contextsect

Detecting installed AI coding agents...

  ✓ Kiro CLI
  ✓ Claude Code
  ✓ Cursor

Select optimization profile:

  1) conservative    — Zero risk. Full exploration.
  2) balanced ⭐     — Recommended. Significant savings.
  3) aggressive      — Maximum savings. Tight budgets.
  4) ultra-aggressive — Absolute minimum tokens.

  Choose profile [1-4, default=2]: 2

  ✓ Profile: balanced

Installing for 3 agent(s) with profile 'balanced'...

  ✓ Kiro: steering + skills + hooks
  ✓ Claude Code: ~/.claude/CLAUDE.md
  ✓ Cursor: .cursor/rules/*.mdc

════════════════════════════════════════════════════════
  ✅ Installation complete!
════════════════════════════════════════════════════════

Update

contextsect update

Switch profile

contextsect profile aggressive

Status

contextsect status