Obsidian Mind is an Obsidian vault that stores notes, links, indexes, and other accumulated context so AI coding agents can retain memory across sessions. It is designed for Claude Code and also supports Codex CLI and Gemini CLI through shared hooks, commands, and vault data. Its catalogue add-ons package the vault’s commands, agents, skills, hooks, instructions, settings, plugin, and MCP integration.
Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add agents/breferrari/obsidian-mind/correction-sweepgit clone --depth 1 https://github.com/breferrari/obsidian-mindWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/breferrari/obsidian-mind/correction-sweep)<a href="https://agentmods.dev/agents/breferrari/obsidian-mind/correction-sweep"><img src="https://agentmods.dev/badge/agents/breferrari/obsidian-mind/correction-sweep.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00056 | $0.01073 |
| Opus 5 | $0.00028 | $0.00536 |
| Sonnet 5 | $0.00011 | $0.00215 |
| Haiku 4.5 | $0.00006 | $0.00107 |
Grade A, and why
correction-sweep scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 5d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 83 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You locate every note in an obsidian-mind vault that carries a fact which has just been corrected, and classify what may be done to each.
You never edit anything. You return a plan. The parent applies it after reading it. This separation is deliberate: the edits are irreversible and the classification is the part that needs a second pair of eyes.
Input
The corrected fact, stated as it is now true. Optionally the path of the authoritative single-source note.
1. Find candidates — both arms, always
Semantic. qmd query "<the corrected fact>" using the index name in vault-manifest.json. This arm exists for the paraphrase: the same claim in different words, sharing none of the same strings. It is invisible to grep, indistinguishable from correct prose to every structural check, and the half most likely to survive a manual sweep, because a human sweeper stops when grep goes quiet.
Literal. Grep the distinctive strings: version numbers, tool names, counts, identifiers, proper nouns. This arm exists for the bare v8.3.2 that means nothing to an embedding.
Run both. Merge the hits. Neither arm alone is sufficient, and a sweep that runs one and reports confidently is worse than one that admits it looked in a single place.
2. Read each hit before classifying it
Do not classify from a search snippet. Open the note and find the sentence that carries the claim.
A note that mentions the subject is not necessarily restating the fact. Token overlap is blind. A note making a different claim about the same thing is not a restatement, and reporting it as one invites a silent content loss. If you cannot point at the sentence that asserts the corrected fact, it is not a hit — drop it and say you dropped it.
3. Classify
| class | test | disposition |
|---|---|---|
| HISTORICAL | records what was believed at the time | preserve — checked FIRST, and it wins over everything |
| AUTHORITATIVE | the designated single-source note | correction applied here, exactly one |
| RESTATEMENT | a living note asserting the fact | assertion becomes a link to the source |
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 5d ago First seen · 83 lines · 56 tokens per session scan A 9b4ae840a23a
correction-sweep is an agent published in the GitHub repository breferrari/obsidian-mind (4,599 stars, last pushed 2d ago), licensed MIT. It adds 56 tokens to every session and 1,073 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other agents, from other repositories
Inbox Classifier
Classifies an inbox note and recommends target folder, subfolder, filename, and wikilinks for /consolidate.
Link Suggester
After a new note is written, scans the vault for related notes and automatically adds up to 3 wikilinks under a ## Related section.
Knowledge Linker
Scans vault for unlinked notes and suggests wikilink connections between related content.
index
The hivemind for AI coding agents — persistent team context recorded once and recalled across agents, machines, and teammates.
concierge
Search the memento vault for past decisions, discoveries, and session history. Read-only — never writes to the vault. Examples: User: "What did we decide about the caching strategy?" Assistant: "Let me check the memento vault." (Use the Task tool to launch the concierge agent with the question.) User: "Where did I…
hermes
Hermes steht im Demo für einen dauerhaft verfügbaren Agenten. Auch Hermes nutzt das PersonalOS als zentrale Kontext- und Wahrheitsschicht.