DefuddleSAFE
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: defuddle description: Plan and, with explicit network consent, use an optional external Defuddle cleaner to extract article-like HTTPS pages as Markdown. Use for defuddle, clean this URL, strip page clutter, readable Markdown from a web page, or preparing a web source for later wiki ingestion. --- # Defuddle Treat Defuddle as an optional external extractor, not an internal capability. Cleaning, raw capture, and wiki ingestion are separate operations. Resolve the installed product root from this skill's own location, never from the selected vault or process 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 `$PRODUCT_ROOT`, never relative to the selected vault's `wiki/` directory. If a separately installed `kepano/obsidian-skills` `defuddle` skill is available, prefer it for current CLI flags. Retain the privacy, consent, and transaction rules in this skill. ## Safety contract - Accept remote inputs only as HTTPS URLs. - Reject credentials in URLs, fragments, private or local hosts, non-public IP addresses, control characters, and sensitive query parameters. - Never interpolate a URL into a shell string. Pass it as one argv element. - Treat redirects to a different host as denied until that host is explicitly approved. - State that the URL and request metadata will leave the machine. Network access requires explicit consent in the current request or a separate confirmation. - Do not install a cleaner, execute a placeholder runner, or silently switch to another network fetcher. - Do not claim a fixed token reduction or extraction quality. Inspect the actual output. ## Plan first Create an inert URL plan. This validates the URL and executes no network call: ```bash python3 "$CORE" capture external-plan url "HTTPS_URL" ``` Report the normalized host, network egress, redirect policy, optional external dependency, and `execute: false`. Then inspect whether this package can find a configured Defuddle executable: ```bash python3 "$CORE" contracts --verify --capability defuddle --vault VAULT ``` An `available` state means no executable was found; stop with the inert plan. A `configured` state means the executable was discovered, but this package has no bundled behavioral verifier for it. Show that state and reason, identify the resolved executable path, and require manual review of its provenance, version, and exact argv before execution. Never relabel `configured` as `verified`. When unavailable or when manual review is declined, offer these honest fallbacks: let the user install/configure an external runner, accept a local HTML or Markdown file in `inbox/`, or leave the URL queued for later. Do not claim that content was cleaned, captured, or ingested. ## Execute after consent After network consent, configured-state detection, and manual executable review, invoke the approved executable with an argv equivalent to: ```text defuddle parse HTTPS_URL --md ``` Capture bounded stdout in a temporary draft outside shared vault state. Fail closed on a non-zero exit, empty output, unexpected binary output, an unapproved redirect, or a response that is clearly an authentication/error page. Preserve headings, links, code fences, tables, quotations, and source wording; do not invent missing content. Preview the cleaned Markdown and report extraction limitations. If the user asked only to read or analyze it, keep the result transient. ## Optional raw capture When the user asks to retain the cleaned source: 1. Resolve the user vault. 2. Hash the exact cleaned bytes with SHA-256. 3. Draft a new immutable payload such as `.raw/captured/<sha256>.md` and, when provenance metadata is needed, a new create-only sidecar with the normalized URL, retrieval date, extractor name and version, and content hash. 4. Use `expected_hashes: null` and `mode: create`. If that content-addressed payload already exists with the same bytes, report a no-op; never overwrite it. 5. Build, inspect, and apply one `claude-obsidian.transaction.v1` capture bundle as described in [operation-transactions.md](../wiki/references/operation-transactions.md). Do not create wiki pages, update indexes, assess claims, or mark the source as ingested. Invoke `wiki-ingest` as a distinct requested operation if the user wants the captured payload incorporated into the knowledge base.
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__defuddle.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 Defuddle 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 Defuddle 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 Defuddle access on my machine?
The audit observed no filesystem, network or shell use at all in its source.
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.