Karpathy LLM Memory Wiki
v1.0.9Compilation-over-retrieval knowledge wiki for OpenClaw agents. Drop sources in, get structured cross-referenced pages out. Knowledge compounds instead of dis...
⭐ 0· 30·0 current·0 all-time
byAlex Dobrin@dobrinalexandru
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Benign
medium confidencePurpose & Capability
The name/description match what the files implement: scaffolding for an LLM-maintained wiki (ingest, query, lint) rooted in workspace directories. Required capabilities (creating files under the workspace, optionally scheduling crons or adding the index to a sync script) are proportionate to the stated purpose.
Instruction Scope
Runtime instructions confine reads/writes to workspace locations (work/wiki-sources, memory/wiki, index/log, ingestion-queue). The ingest step may fetch web sources ('fetch URL if web source') which is reasonable for ingesting remote docs but means the agent can fetch arbitrary external URLs when asked to ingest. The setup script appends a one‑time Wiki reference to MEMORY.md during install; otherwise the SKILL explicitly declares exclusion zones it will not modify.
Install Mechanism
This is instruction-only with a local bash setup script and template files — no network downloads, no package installs, no archives extracted. The script copies templates into your workspace and optionally invokes the openclaw CLI to create cron jobs. This is a low-risk install mechanism if you review the script first.
Credentials
No credentials or secret environment variables are requested. The script respects OPENCLAW_WORKSPACE_DIR if present and uses HOME/openclaw.json when present. There are no unexplained secret requests.
Persistence & Privilege
always: false. The skill is user-invocable and can be run autonomously by the agent (platform default). Creating scheduled crons via openclaw cron create (opt-in) will give the agent periodic autonomous runs (ingest/lint). Enabling SuperMemory sync (--with-sync) will add the wiki index to a cloud sync script if you opt in; that is an explicit, opt-in persistence choice. Be mindful that opt-in cron + autonomous invocation increases the chance of repeated agent actions if you enable them.
Scan Findings in Context
[meta_version_mismatch] unexpected: _meta.json and registry metadata record different versions (1.0.6 vs registry 1.0.9). This is a packaging/metadata inconsistency and not expected for a clean publish, though not itself malicious.
[origin_author_mismatch] unexpected: .clawhub/origin.json lists author/slug 'dobrinalexandru/wiki-system' whereas registry owner ID and slug in the package metadata differ. This suggests the package may have been repackaged or forked; it merits extra caution (or verification of provenance) before blindly trusting.
Assessment
This skill is functionally coherent for building an LLM-maintained wiki, but review a few points before installing:
- Inspect scripts/wiki-setup.sh yourself (it will copy templates into your workspace and append a one‑time Wiki reference to MEMORY.md if present). Run it only in a workspace you control.
- The ingest operation can fetch external URLs when asked — only add trusted sources to work/wiki-sources/ingestion-queue.md.
- Cron creation and SuperMemory sync are opt‑in. Do not pass --with-cron or --with-sync unless you want scheduled autonomous agent runs and/or cloud sync of the wiki index.
- Metadata mismatches (package version and origin author/slug) suggest the package may have been repackaged; verify the source or author if provenance matters to you.
- If you want minimal risk, run the setup script without cron/sync, keep work/wiki-sources local, and manually invoke 'wiki ingest' when you control the sources. If you enable cron or cloud sync, restrict the workspace and review what will be synced to the cloud.Like a lobster shell, security has layers — review code before you run it.
latestvk97019vcn7a4dzw8b2n07tnhnh84c9xg
License
MIT-0
Free to use, modify, and redistribute. No attribution required.