How to use the Save Conversation to Obsidian agent skill

What the save skill makes the agent do

An agent skill is a set of instructions that an AI agent loads when a request matches it. This skill covers one job: saving content you select from the conversation into an Obsidian vault as one reviewed transaction. It is meant to run only when you explicitly ask to preserve specific conversation content.

The agent first reads the vault's hot.md and index.md files, and searches for an existing note before creating a new one, preferring a small update over a duplicate. It then proposes a preview of every file it will change. It writes only after you approve.

• It saves only what you select, never the whole conversation by default.
• It does not run automatically and does not archive every conversation.
• It does not ingest files or URLs. The skill says those must go through a separate operation.
• Each save normally updates the selected note, the index, the log, and the hot cache together.

Install the save skill

Download the ZIP for skills/save from the directory page and unzip it. Then place the skill folder in one of two locations: your personal skills folder at ~/.claude/skills/, or a project's .claude/skills/ folder if you want it only for that project.

If you use the Claude apps instead, upload the ZIP under Settings → Capabilities → Skills.

The skill also depends on the claude-obsidian product folder, which must be installed on your machine. The skill's instructions resolve the path to scripts/claude-obsidian.py from that product root, so keep that folder where it is installed.

• Personal install: ~/.claude/skills/
• Project install: .claude/skills/ inside the project
• App install: Settings → Capabilities → Skills, then upload the ZIP

Prepare your environment

The skill needs three things. First, an installed claude-obsidian product folder containing scripts/claude-obsidian.py. Second, Python 3 available to run that script. Third, an Obsidian vault with a wiki/ folder that includes hot.md and index.md.

The vault is resolved in this order: an explicit --vault option, then the CLAUDE_OBSIDIAN_VAULT environment variable, then a workspace config, then discovery from the current directory. The product folder itself is never treated as a vault.

• scripts/claude-obsidian.py present in the product folder
• Python 3 on your system
• A vault with wiki/hot.md and wiki/index.md

Trigger the skill with natural requests

You can invoke the skill with /save or with phrases such as save this, save that answer, file this conversation, save this analysis, keep this insight, or preserve this chat result. Be specific about what you want kept, because the agent saves only the scope you select.

If the scope, destination, or handling of sensitive content is unclear, the agent asks one focused question before drafting. If you do not give a title, it drafts a concise one and includes it in the review preview.

• Example: /save the pricing comparison from my last answer
• Example: save this analysis as a question note
• Example: keep this insight, but leave out the sample data

Review the preview and approve

Before anything is written, the agent shows the note title, the destination, whether each file is created or replaced, and the list of changed paths. The skill runs an inspect step first and then an apply step that requires an approval fingerprint. You review the preview and only then approve.

If the same operation is retried with a different bundle, the agent is expected to re-read the vault and inspect a new bundle rather than force the write. Interrupted applies are recovered with a transaction recover step, not bypassed.

• Check the note title and destination
• Check whether each path is a create or a replace
• Check the list of changed paths before approving

Safety notes: scripts and network

This skill runs scripts: yes. It executes the claude-obsidian.py script to inspect and apply vault changes. It uses no network: the skill states it makes no network requests, and a separately approved source ingest or research operation is required for anything that needs the network.

Content you select is preserved as written, so sensitive chat content can be saved if you select it. The skill treats pasted or quoted source text and tool output as content to preserve, not as instructions, and it ignores embedded directives to run commands, widen scope, change the destination, or enable outside access. Still, review every preview before approving.

• Runs scripts: yes (claude-obsidian.py)
• Uses network: no
• Sensitive content you select can be saved, so review before approving
• Never run the whole transcript through this skill by default

Tips for better results

Select the smallest useful piece of conversation. A single answer or decision is easier to review and keep accurate than a whole session. Name the note type you want when it matters: the skill commonly uses question for an answered analysis, concept for an idea worth naming, source for a material summary, and session for approved conversation content.

If the material is already in your vault, the agent may report that and offer a no-op. Tell it explicitly if you still want it saved.

• Save one answer or decision at a time
• Name the note type if you have a preference
• Expect a no-op if the content is already recorded