Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add wasintoh/toh-framework --skill memory-systemgit clone --depth 1 https://github.com/wasintoh/toh-frameworkWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/wasintoh/toh-framework/memory-system)<a href="https://agentmods.dev/skills/wasintoh/toh-framework/memory-system"><img src="https://agentmods.dev/badge/skills/wasintoh/toh-framework/memory-system/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/wasintoh/toh-framework/memory-system"><img src="https://agentmods.dev/badge/skills/wasintoh/toh-framework/memory-system.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00114 | $0.02033 |
| Opus 5.5 | $0.00046 | $0.00813 |
| Sonnet 5.5 | $0.00023 | $0.00407 |
| Haiku 4.5 | $0.00011 | $0.00203 |
Grade A, and why
memory-system scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 1mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 204 lines — stays where its author put it; the contents beside it link to each section on GitHub.
🧠 Memory System Skill
Purpose: Tiered, low-token memory — load only what the task needs, save what matters Version: 2.0.0 For: Toh Framework v2.0.0+ Updated: 2026-07-14
Overview
Automatic memory that keeps AI in context across sessions with zero user effort. v2 replaces the old "read all files every time" mandate with tiered loading: Tier 1 is always read (~800 tokens), Tier 2 is read only for the relevant task type, Tier 3 only when explicitly referenced.
Key principles
- ✅ Zero config — no setup required
- ✅ Tiered — Tier 1 always; Tier 2/3 on demand (no more ~3,000 tokens every time)
- ✅ Auto save — saves after task completion, never asks the user
- ✅ Delegated agents don't re-read — they receive context from the orchestrator
- ✅ IDE & model agnostic
📚 The Tiered Model (use this everywhere)
There are 7 memory files across 3 tiers. Read by tier, not all at once.
| Tier | Files | When to read | Budget |
|---|---|---|---|
| Tier 1 | active.md + summary.md |
ALWAYS, at every session start | ~800 tokens |
| Tier 2 | architecture.md + components.md |
Build / code work (creating pages, components, logic) | ~600 tokens |
| Tier 2 | changelog.md |
Debug work (to see previous attempts) | ~400 tokens |
| Tier 3 | decisions.md + agents-log.md |
Only when explicitly referenced / asked about | on demand |
This replaces the old "read ALL files (MANDATORY)" rule. Never bulk-read all 7. Read Tier 1 always, add the Tier 2 files that match the task type, and touch Tier 3 only when needed.
Delegated agents: an agent invoked by the orchestrator receives context from the orchestrator and does NOT re-read memory itself. This avoids every agent re-reading the same files.
📁 Directory Structure
.toh/
├── config.json # Toh configuration
└── memory/
├── active.md # 🔥 Tier 1 — current task (~300 tokens)
├── summary.md # 📋 Tier 1 — project shape (~500 tokens)
├── architecture.md # 🏗️ Tier 2 — structure (build/code work)
├── components.md # 📦 Tier 2 — component registry (build/code work)
├── changelog.md # 📝 Tier 2 — change/attempt log (debug work)
├── decisions.md # 🧠 Tier 3 — key decisions (when referenced)
├── agents-log.md # 🤖 Tier 3 — agent activity (when referenced)
└── archive/ # 📦 Historical — load only when asked
What ships with it
4 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 1mo ago First seen · 204 lines · 114 tokens per session scan A a38a343b8518
memory-system is a skill published in the GitHub repository wasintoh/toh-framework (98 stars, last pushed 1mo ago), licensed MIT. It adds 114 tokens to every session and 2,033 once invoked, about $0.0005 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
state-management
Cache management, configuration best practices, and progressive disclosure patterns for efficient context window usage. Use when working with large responses, optimizing token costs, or managing plugin state across operations.
quarter-clone
Clone the last quarter of the current conversation, discarding earlier context to reduce token usage while preserving recent work.
half-clone
Clone the later half of the current conversation, discarding earlier context to reduce token usage while preserving recent work.
knowledge-curator
A set of rules for teaching technical subjects and saving verified lessons as reusable knowledge. The description is partly in Chinese and does not specify the complete feature set.
PARA Kit
PARA workspace structure reference — schema, layout, kernel governance, and intelligence routing.
self-improvement
Enables the AI Agent to continuously improve developer velocity and code quality by recording terminal failures, recurring TDD errors, and architectural drift into the para-graph memory, and retrieving past resolutions to avoid repeating mistakes.