Who it is forA team lead or careful developer who does not want the model to write facts into memory without review.
The problemA model may save incorrect or unwanted facts to long-term memory without anyone noticing.
The outcomeEvery proposed write is held for approval, and a denied write leaves a *-denied audit row instead of changing memory.
What you need first
- Node ^22.19.0 or >=24.0.0
- DeepSeek Harness with the dsh CLI installed
- A dsh profile, such as the web profile used in the examples
Steps
- Run
dsh plugin --profile web add dsh-mementoto install the plugin. - Run
dsh --profile web --dump-config | grep -A3 'id: memento'and confirm writePolicy is ask. - Trigger a memory write in a session and deny it in the approval prompt.
- Run
/memory auditand find the *-denied row for that write. - Run
/memory listand confirm the denied entry is absent.
Prompt for your agent
In the dsh-memento repo, confirm writePolicy is ask, trigger one memory write from <session>, deny it, then run /memory audit. Expected output: a *-denied audit row for that write. Verify that /memory list does not show the denied entry.
Caveat. The gate only covers writes that go through the plugin service, and the README does not describe the approval prompt layout, so check the exact flow in your own harness build.
Use this in your AI agent
Ask your agent
Using the Deduped MCP, get the setup steps for PerryLink/dsh-memento and "Review every memory write before it is saved"
Add Deduped to Claude Code once
claude mcp add --transport http deduped https://dedupedai.com/mcp
Every Monday: the 5 hottest repos and exactly how to use them. Free, one email a week, unsubscribe anytime.
Get them freeShare: X LinkedIn Hacker News Email Copy link
Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer