Wikimind Skill

Ingest articles, docs, notes, and web pages into your local LLM-WikiMind knowledge base (Karpathy's LLM Wiki pattern). Triggers: add to knowledge base, inges...

Security Scan

The skill's requirements and runtime instructions are coherent with its stated purpose of writing Markdown pages into a local WikiMind knowledge base; nothing requested appears unrelated or excessive.

Purpose & Capability

Name/description, required binaries (python3, qmd), and the runtime steps (write Markdown files, update index with qmd, append to a log) all align with an ingest-to-local-wiki workflow. The mention of a local MCP 'wiki-kb' server is consistent with the preferred ingestion API.

Instruction Scope

Instructions are narrowly scoped to reading the wiki schema, classifying content, writing a Markdown file under $WIKIMIND_ROOT, running 'qmd update', and appending to a local log. Two minor notes: (1) the SKILL.md references the environment variable $WIKIMIND_ROOT but the skill metadata does not declare it; (2) the preferred 'wiki_ingest_note(...)' call assumes an MCP API is available but implementation details are not included. Neither is a direct security red flag, but they are important operational details the user should verify.

Install Mechanism

This is an instruction-only skill with no install spec and no code files, so nothing is downloaded or written by the skill itself. Lowest-risk installation surface.

Credentials

The skill declares no required credentials or config paths, which is appropriate. However, it will read/write files under $WIKIMIND_ROOT (an environment variable referenced but not declared) and expects a registered 'wiki-kb' MCP server in the platform — the user should confirm what credentials or agent/server registrations that MCP requires. The skill does not request unrelated secrets.

Persistence & Privilege

always:false and user-invocable:true (normal). The skill does not request permanent/automatic inclusion or attempt to modify other skills or system-wide settings.

Scan Findings in Context

[no-findings] expected: Regex-based scanner found nothing; this is expected because the package is instruction-only (no code files) and its runtime behavior is expressed in SKILL.md.

Assessment

This skill appears to do what it says: write standardized Markdown pages into a local WikiMind repo and update the search index. Before installing, confirm: (1) where your wiki root is (the skill uses $WIKIMIND_ROOT — set and verify this path, and consider using a test wiki first); (2) that 'qmd' is installed from a trusted source (pip3 install qmd) and works on your system; (3) that the 'wiki-kb' MCP server you register in CatDesk/OpenClaw is trusted and you understand any credentials it may store; (4) you are comfortable allowing the agent to create/modify files under the wiki root (backup important content). If you want stricter control, run the skill only when explicitly invoked rather than allowing autonomous invocation.

wiki-ingest Skill

Ingest content into your local LLM-WikiMind knowledge base — a production implementation of Karpathy's LLM Wiki pattern.

Prerequisites

1. LLM-WikiMind installed — see setup guide
2. wiki-kb MCP server registered in CatDesk/OpenClaw
3. qmd installed: pip3 install qmd

Trigger phrases

• "Add this to my knowledge base"
• "Ingest this article"
• "Save to wiki" / "Write to knowledge base"
• "Store this note"
• wiki_ingest
• 加到知识库 / 写入知识库 / 保存到 wiki / 记录到知识库 / 把这个存起来

Workflow

Step 1: Read global schema

cat "$WIKIMIND_ROOT/CLAUDE.md"

Step 2: Classify the content

Type	Directory	When to use
concept	concepts/	Explaining a concept, pattern, or mechanism
entity	entities/	Describing an API object or class
comparison	comparisons/	Comparing two approaches
source-summary	sources/	Summarizing an article or doc

Step 3: Identify the domain

• Use an existing domain (check CLAUDE.md for the list)
• Or create a new one: make a directory + DOMAIN.md with keywords frontmatter

Step 4: Write the page

Preferred — use the MCP tool (if wiki-kb server is loaded):

wiki_ingest_note(
  title="Page Title",
  content="# Page Title\n\nContent here...",
  domain="my-domain",
  page_type="concept",
  source="https://example.com",
  summary="One-line summary",
  tags=["tag1", "tag2"]
)

Fallback — write the file directly:

---
title: "Page Title"
type: concept
domain: my-domain
source: "https://example.com"
summary: "One-line summary (<=150 chars)"
tags: ["tag1"]
related: []
created: YYYY-MM-DD
updated: YYYY-MM-DD
confidence: medium
---

# Page Title

Content here.

Save to: $WIKIMIND_ROOT/<domain>/concepts/page-title.md

Step 5: Update the search index

cd "$WIKIMIND_ROOT" && qmd update

Step 6: Append to log

echo "## [$(date +%Y-%m-%d)] ingest | Page Title | domain" >> "$WIKIMIND_ROOT/<domain>/log.md"

Quality Standards

concept pages must include: definition, motivation, code example, common pitfalls, related links

entity pages must include: description, key properties with types, key methods with signatures, usage example, gotchas

comparison pages must include: comparison table, when to use each, migration notes

source-summary pages must include: original URL, key takeaways (3-7 points), what this adds vs existing knowledge

Notes

• Never write into refs/ — that's for bulk-imported raw docs
• confidence: high for official docs, medium for your summaries, low for speculation
• After writing, tell the user the file path and a good search query to find it