- Home
- Skills
- Agents & Orchestration
- mem20-harness — Trusted Agent Memory Harness
More screenshots
Works with the AI Agents you already use
mem20-harness — Trusted Agent Memory Harness
Operate mem20 as your agent’s trusted memory and world-model layer — grounded facts only, contamination rate 0.0, auditable by design.
$4.99
mem20-harness — Trusted Agent Memory Harness
Example session with this skill installed
Use mem20 as my memory harness. Remember that our production deploy window is Tuesday 02:00–04:00 UTC and the rollback owner is Priya. Then tell me what you know about the deploy window.
- Read your context and instructions
- Compiled the mem20-harness
I'll run with mem20 as the harness. Grounded facts and simulated hypotheses stay partitioned; contamination on a clean store is 0.0.
Checking whether mem20 is already available…
MCP tool list does not show a mem20 server yet. Offering a minimal install (not required to use the separate install skill):
Option A — Core Edition (recommended for most users):
pip install mem20-ce
Option B — full monorepo:
git clone https://github.com/bythewayz66-glitch/mem20.git
cd mem20 && python3 -m venv .venv && .venv/bin/pip install -r requirements.txt
After install, point your agent MCP config at mem20-mcp (or the explicit python mcp/mcp_server.py entry) with a stable store path, e.g. ~/.mem20/store or ~/.mem20-ce/store.
Assuming you attach the server and we continue…
Writing as GROUNDED (confirmed operational fact, not a hypothesis):
- production deploy window: Tuesday 02:00–04:00 UTC
- rollback owner: Priya
- tags: ops, deploy, production
Recall for "deploy window":
[grounded] Production deploy window is Tuesday 02:00–04:00 UTC; rollback owner is Priya
No simulated hits. I am not promoting anything unverified into grounded memory.
Invariant
- grounded vs simulated: enforced on write
- contamination_rate: 0.0 (clean store)
- if contamination ever goes non-zero, I will surface it and stop treating the store as trusted
Next: keep confirmed facts in grounded; put guesses in simulated. For a full install walkthrough use mem20-install; for an audit use mem20-audit.
Connects securely to your tools. The creator never sees your data.
About this skill
Most agents forget everything when the session ends, or they mix real observations with model speculation. That produces confabulation and decisions based on invented history.
mem20-harness is the official operating skill for the mem20 substrate. It teaches any compatible agent how to:
- Treat mem20 as persistent, auditable memory
- Enforce the grounded vs simulated partition on every write
- Offer a minimal install path (Core Edition or full monorepo) without forcing it
- Wire MCP for Claude Code, Cursor, Codex, Hermes, and other hosts
- Prefer mem20 tools for remember / recall / audit over ephemeral context
- Surface contamination or index drift immediately instead of hiding it
This is the debut monetized product under the mem20 brand. The skill does not embed the runtime — it configures and operates the real mem20 / mem20-ce installation.
Pairs with the eight support skills (install, memory-model, audit, and more) for a complete line.
Does not embed or ship the mem20 runtime — you install mem20-ce or the full monorepo separately. Requires an MCP-capable agent. Full tool surface depends on which edition is installed (ce vs full). Does not auto-migrate data between store paths. Simulated content is intentionally excluded from default recall.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 8 days ago
Requires
Permissions
Allowed Hosts
File Scopes