curate

A review tool for pending memory notes that have been submitted to a shared knowledge vault but not yet checked against its existing notes.

In plain words
What is it for?
Use it when new memory offers are waiting for review, after a memory write signals pending notes, or when you are explicitly asked to curate the vault.
Why use it?
It helps prevent duplicate, overlapping, or conflicting lessons from entering the vault without review. The notes are judged locally against the vault's current contents.

Skill for Claude CodeCodex

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 skills/toejough/engram/curate
Any agent
npx skills add toejough/engram --skill curate
Clone the repo
git clone --depth 1 https://github.com/toejough/engram

Made for: Claude Code, Codex.

Per session 99 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 1,338 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 $0.00099 $0.01338
Opus 5 $0.00049 $0.00669
Sonnet 5 $0.00020 $0.00268
Haiku 4.5 $0.00010 $0.00134

Measured 2d ago against content hash 692ec5f5be46, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

curate 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.

agent-instructions/skills/curate/SKILL.md · 84 lines

How it starts

The opening of the file, as written. The whole thing — 84 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Curate — Judge Pending Offers Against the Host Vault

A served engram learn/engram amend write lands as a pending offer: a real note file, marked pending: true, that hasn't yet been reviewed against the vault's existing (non-pending) notes. Curation is that review — the same agent-judged covered/near/absent reasoning recall's Step 2.5 performs against query candidates, applied instead to a pending offer against the host vault's existing notes.

Host-local only, on your own initiative. engram serve's only job on a write is authenticate, stamp identity, persist the pending marker, respond — judgment happens later, off the request path, never synchronously inside the HTTP request that created the offer. Invoke this skill reactively, whenever you notice one of the three surfacing signals (query payload flag, update notice, write-path nudge), or whenever explicitly asked to curate.

Step 1 — Find pending offers

engram query excludes pending offers from its results by design — it can only tell you THAT some exist (pending_offers: true in its payload), not which ones. There is no CLI listing for them (engram count --group-by pending only counts). Scan the vault directly instead:

grep -l '^pending: true$' <vault>/*.md

Read each match in full (frontmatter + body) — that's the offer's claim.

Step 2 — Judge each offer against the host vault's existing notes

engram query --phrase "<offer's situation>" finds related existing notes the normal way (it already excludes offers, so every result is a real candidate to judge against). For each offer:

Outcome Criterion Action
Covered an existing note already states the offer's claim, no material omission reinforce it: engram amend --target <existing> --activate --chunk-source <ids> (carry forward any chunk-source the offer already had) [--supersedes ... if it corrects a different, outdated note] — then discard the offer: engram amend --target <offer> --discard
Near overlaps an existing note's topic but adds ≥1 substantive claim the existing note omits fold the new claim in: engram amend --target <existing> --chunk-source <ids> --subject/--predicate/--object (or --behavior/--impact/--action) [--supersedes ... if correcting] — then discard the now-redundant offer: engram amend --target <offer> --discard
Absent no existing note addresses the offer's situation accept as-is, clear its own marker: engram amend --target <offer> --clear-pending

Read the full file on GitHub · 84 lines

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 · 84 lines · 99 tokens per session scan A 692ec5f5be46

Subscribe to this mod's changes

curate is a skill published in the GitHub repository toejough/engram (8 stars, last pushed 2d ago), licensed Apache-2.0. It adds 99 tokens to every session and 1,338 once invoked, about $0.0005 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.