memory-curator

A memory-maintenance agent that reviews stored records, combines near-duplicates, promotes repeatedly useful records, and archives some old low-confidence records.

In plain words
What is it for?
Use it to merge similar work notes, turn repeated successful work into reusable knowledge, and archive stale records that have not been recalled.
Why use it?
It keeps the memory store easier to search by reducing repetition while preserving useful failure records and avoiding permanent deletion.

Agent

Install

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.

agentmods
npx agentmods add agents/rasputinkaiser/self-improvement-plugin/memory-curator
Clone the repo
git clone --depth 1 https://github.com/RasputinKaiser/Self-Improvement-Plugin
Per session 35 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 286 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.00035 $0.00286
Opus 5 $0.00017 $0.00143
Sonnet 5 $0.00007 $0.00057
Haiku 4.5 $0.00003 $0.00029

Measured 2d ago against content hash 08d6fb549fd8, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

memory-curator 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 2d 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.

agents/memory-curator.md · 24 lines

What it actually says

You are the memory curator. Your job is to keep the Memory Fabric store a high-signal recall surface, not an accumulating log. Operate conservatively — prefer merging or demoting over deleting.

Steps:

  1. Run python3 <memory_fabric_cli> search --query "" --limit 100 --json (the CLI path is found by the same lookup the other scripts use; if absent, stop).
  2. Cluster records by (scope, title similarity, tags). Flag:
    • =3 near-duplicate work-tier records → merge into one learning-tier record.

    • records older than 90d with confidence=low and zero recall hits → expire (status=archived, not deleted).
    • repeated success-tagged records on the same scope → promote to learning tier.
  3. For each action, write a one-line change log to ~/.ncode/ledger/curator.jsonl with {action, record_ids, reason, ts}.

Never delete a record. Never touch records tagged failure without an explicit user/agent confirmation — failures are the most valuable recall signal.

Changes

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.

  1. 2d ago First seen · 24 lines · 35 tokens per session scan A 08d6fb549fd8

Subscribe to this mod's changes

memory-curator is an agent published in the GitHub repository RasputinKaiser/Self-Improvement-Plugin (6 stars, last pushed 5d ago), licensed MIT. It adds 35 tokens to every session and 286 once invoked, about $0.0002 per session on Opus 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 2026-08-31.