Claude External Context System

Claude That Actually Follows Rules

The first system that doesn't just store context for Claude — it enforces it. Built from real trading sessions. Zero installation. Works in claude.ai chat.

Free on GitHub → Get Premium Templates
R2,297
Profit — Week 1
0
Rules Violated After Setup
<10min
Setup Time
13
Proven Trading Rules
The Problem

Claude Reads Rules.
It Doesn't Follow Them.

Every Claude memory solution stores your rules as passive text. Claude reads them — sometimes. Under pressure, during fast analysis, when things get complex — it skips steps. That's how a R1,046 account nearly got wiped on one trade.

01

Internal memory fails

Claude's built-in memory stores rules as suggestions. Process steps get skipped when Claude is under pressure to respond quickly.

02

Existing solutions require installation

claude-mem, claude-memory, supermemory — all require npm, Python, terminals, and background processes. Not for regular users.

03

Memory ≠ enforcement

Remembering a rule and being blocked from breaking it are completely different things. No existing system enforces — they only remind.

04

Context resets every conversation

Every new chat, Claude starts from zero. Months of learned behaviour, rules, and lessons — gone. You re-explain everything.

The Enforcer In Action

Bad Trades Get Blocked.
Automatically.

// Claude attempts to place a Gold trade at 3 units...

=== PRE-TRADE ENFORCER ===
Instrument: XAUUSD | Direction: Buy | Entry: 4,461
SL: 4,442 (19pts) | TP: 4,490 (29pts) | Size: 3 units

1. News scan done?
2. Correct session?
3. 5+ instruments scanned?
4. H4 confirms trend?
5. H1 pullback confirmed?
6. M15 entry signal?
7. Lot size correct? ❌ — 3 units = R48/pt = R912 risk
8. SL behind structure?
9. R:R 1.2:1+?
10. Risk under 20%? ❌ — R912 = 44% of account

RESULT: FAILED — ITEMS 7 AND 10 — TRADE BLOCKED
// Fix: Reduce to 1 unit maximum on Gold. Rerun enforcer.
Why Claude ECS

Built Different.
Proven Different.

Feature Other Solutions Claude ECS
Works in claude.ai chat Claude Code only Any interface
Zero installation npm / Python required Nothing to install
Rule enforcement Passive memory only Enforcer blocks violations
Human readable SQLite / vector DBs Plain markdown file
User can edit directly Technical skill required Edit on GitHub in browser
Verification protocol Not available Confirmation phrase
Pre-action checklists Not available Built-in enforcer
Portable across accounts Tied to local machine Any Claude, anywhere
Setup time Hours to days Under 10 minutes
Pricing

Start Free.
Scale When Ready.

Free Forever
Community
R0
Open Source — MIT License
  • Master prompt (full)
  • Basic trading template
  • Basic dev template
  • GitHub community access
  • Community lessons & rules
Get on GitHub →
Developers
Dev Pro
R299
One-time payment
  • Full dev context template
  • Pre-code enforcer checklist
  • Bug tracking system built in
  • Architecture decision log
  • Sprint planning protocol
  • Code review checklist
  • 30 days email support
Get Dev Pro →

Enterprise implementations available — contact for pricing. Community tier is and will always remain free.

EVAN
GROB
BELAAR
Evan Grobbelaar
Creator — Claude ECS | Johannesburg, South Africa | 2026

Claude ECS wasn't designed in theory. It was built through real live trading sessions on a R1,046 demo account — with real mistakes, real losses, and real lessons committed to GitHub after every session.

After a near-miss where Claude placed a Gold trade risking R1,042 on a R1,046 account due to wrong lot sizing, the question became: how do we build a system that Claude actually follows — not just reads?

The answer became the External Context System. Week 1 results: +R2,297 profit. Rules violated after system setup: zero.

Evan also builds ClockPay — a biometric time-attendance and payroll system for South African SMEs — and is the founder of the Claude ECS open source community.

GitHub → Contact →