CanvasSAFE
Self-organizing AI second brain for Obsidian + Claude Code. Drop any source and Claude reads, links, and files it into one connected knowledge graph of plain Markdown you own. AI note-taking, personal knowledge management (PKM), and an open-source Notion alternative. Based on Karpathy's LLM Wiki pat
Overview
Self-organizing AI second brain for Obsidian + Claude Code. Drop any source and Claude reads, links, and files it into one connected knowledge graph of plain Markdown you own. AI note-taking, personal knowledge management (PKM), and an open-source Notion alternative. Based on Karpathy's LLM Wiki pat
3ba6310f0e63OBSERVED · 2026-10-07What it tells the agent
The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.
---
name: canvas
description: Create, inspect, and update Obsidian JSON Canvas boards with text, file, link, group, and edge nodes. Use for canvas status, canvas lists, visual maps, zones, spatial layouts, adding vault notes or media to a .canvas file, and requests such as create canvas, add to canvas, or put this on the canvas.
---
# Canvas
Treat a canvas as a vault-scoped JSON Canvas document. Keep reads read-only and
route every requested mutation through one recoverable transaction.
Resolve the installed product root from this skill's own location, not from the
vault or current working directory:
```bash
PRODUCT_ROOT=/absolute/path/to/installed/claude-obsidian
CORE="$PRODUCT_ROOT/scripts/claude-obsidian.py"
test -f "$CORE"
```
Every `../wiki/references/` link in this file resolves the same way, relative
to this skill's own directory under `$PRODUCT_ROOT`, never relative to the
selected vault's `wiki/` directory.
## Choose the syntax source
Prefer a separately installed `kepano/obsidian-skills` `json-canvas` skill for
format details. Otherwise read [references/canvas-spec.md](references/canvas-spec.md).
The open standard is [JSON Canvas 1.0](https://jsoncanvas.org/spec/1.0/).
Use this skill for vault workflow and transaction safety even when an upstream
syntax skill is available.
## Scope
- Resolve the user vault before reading. Never derive it from plugin files.
- Store boards under `wiki/canvases/`; use `wiki/canvases/main.canvas` only as
the default when the user did not name a board.
- Treat `wiki/canvases/index.md` as an optional catalog. Update it only when a
canvas is created, renamed, or removed.
- Use vault-relative paths in `file` and `background`; reject absolute paths,
`..` traversal, home shortcuts, and paths that escape through symlinks.
- Reference only files already inside the vault. If an optional capture or
media capability is detected, compose with it and consume its returned
vault-relative path. Do not assume a particular plugin, command, or session
log, and do not download or copy assets in this skill.
- Use `link` nodes for URLs. Accept HTTPS URLs only; a link node does not grant
permission to fetch the page. Creating its JSON makes no request, but opening
it in Obsidian may fetch Open Graph metadata from that URL's host. Disclose
that render-time egress and confirm it is acceptable; otherwise use a text
node containing the URL.
## Read-only operations
For status or list requests, parse the selected `.canvas` files and report node
counts, group labels, broken edge endpoints, and missing vault-relative file
targets. If the default canvas is absent, report that fact and offer a creation
preview; do not create it during a status request.
## Draft a mutation
1. Read the entire canvas and any catalog target. Record each expected SHA-256,
or `null` for a file that must be absent.
2. Preserve unknown JSON fields and existing array order. The node array is
bottom-to-top z-order.
3. Draft the requested nodes or edges:
- Give every node and edge a unique ID. Prefer a random 16-character
lowercase hexadecimal ID and verify it is unused.
- Require integer `x`, `y`, `width`, and `height` values for every node.
- Use `text`, `file`, `link`, or `group` exactly as defined by JSON Canvas.
- Require every `fromNode` and `toNode` to reference an existing or newly
drafted node.
4. Position new content deliberately. A group is a visual container, not a
parent. Use 20 px inner padding and 40 px gaps, wrap to a new row when
needed, and preview any group expansion.
5. Serialize valid UTF-8 JSON with top-level `nodes` and `edges` arrays.
Minimal node examples:
```json
{
"nodes": [
{
"id": "6f0ad84f44ce9c17",
"type": "text",
"text": "# Research map\n\nA concise orientation card.",
"x": 0,
"y": 0,
"width": 400,
"height": 180
},
{
"id": "a1b2c3d4e5f67890",
"type": "file",
"file": "wiki/concepts/Contextual Retrieval.md",
"x": 460,
"y": 0,
"width": 400,
"height": 240
}
],
"edges": []
}
```
Use `file` plus optional `subpath: "#Heading"` for notes, images, PDFs, and
other vault files. Use `url` only on a `link` node. Escape line breaks in JSON
strings as `\n`, not as literal backslash-plus-`n` text.
## Preview and apply
Validate before showing the preview:
- JSON parses and uses supported node types.
- IDs are unique and edge endpoints exist.
- dimensions are positive integers;
- file and background paths are safe, vault-relative, and present;
- new nodes do not unintentionally overlap or overflow their requested group.
Build one `claude-obsidian.transaction.v1` bundle with operation type `canvas`.
Include the canvas and the catalog in that same bundle when the catalog changes.
Read [operation-transactions.md](../wiki/references/operation-transactions.md),
then inspect the bundle before applying it:
```bash
python3 "$CORE" transaction inspect BUNDLE --vault VAULT
# Set APPROVAL_SHA256 to the inspect result's approval_sha256 after review.
python3 "$CORE" transaction apply BUNDLE --vault VAULT \
--approved-plan-sha256 "$APPROVAL_SHA256"
```
For removals, replacements, renames, or other destructive changes, obtain
explicit consent after presenting the preview. Report the operation ID, exact
changed paths, board name, node IDs, and final positions. Do not commit Git.Trust audit
SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.
| Layer | What it checks | Result |
|---|---|---|
| L0 | Provenance & inventory | PASS |
| L1 | Static analysis of the code | NA |
| L2 | Instruction surface (what it tells the agent) | PASS |
| L3 | Class-specific surface | PASS |
| L4 | Behavioural (sandbox) | SKIPPED |
What the source does
- Filesystem
- none-observed
- Network
- none-observed
- Shell
- none-observed
- Dependencies
- pinned
- Secrets in source
- none-found
Findings (0)
No findings outside the package's declared scope.
Gates applied: no_behavioural_pass.
3ba6310f0e63full audit observations/trust-audit/skill/agricidaniel__canvas.json · Report an issue / request a re-scanAudit history
Every audit this skill has had.
| Date | Source | Verdict | Grade | Score | Change |
|---|---|---|---|---|---|
| 2026-10-07 | 3ba6310f0e63 | SAFE | B | 89 | first audit |
Questions
What does the Canvas skill do?
Self-organizing AI second brain for Obsidian + Claude Code. Drop any source and Claude reads, links, and files it into one connected knowledge graph of plain Markdown you own. AI note-taking, personal knowledge management (PKM), and an open-source Notion alternative. Based on Karpathy's LLM Wiki pat
Is Canvas safe to install?
The audit found nothing in the source that contradicts what it says it does, and graded it B (89/100). Grade A is held back for packages that have also passed a sandboxed behavioural run, which is why a clean skill reads B.
What can Canvas access on my machine?
The audit observed no filesystem, network or shell use at all in its source.
What do I need installed to use Canvas?
Its own instructions reference height and width. Dependencies are pinned to exact versions.
How current is this page?
The grade is for one exact copy of the source (3ba6310f0e63), read on 2026-10-07. The repository is watched, and a new audit runs when it changes — this is the first audit.