Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Second Brain Wiki

    1

    Most "chat with your files" setups rediscover your notes from scratch on every question, so nothing builds up.

    Free

    Security scanned
    Second Brain Wiki

    Second Brain Wiki

    Example session with this skill installed

    I clipped the Wikipedia article on spaced repetition into raw/articles/. Ingest it into my wiki.

    • Read your context and instructions
    • Compiled the second brain wiki

    Ingested

    Spaced repetition (Wikipedia, CC BY-SA 4.0, rev. 1383252775) and connected it to the forgetting-curve material already in your wiki. 9 pages created, 5 updated; lint is clean (17 pages, 264 links, 0 errors). Committed as ingest: Spaced repetition (Wikipedia).

    New pages

    PageWhat's in it
    sources/spaced-repetition-wikipediaFull summary: history, research, algorithms, evidence and criticism, implementations, reliability notes
    concepts/spaced-repetitionDefinition, how expanding schedules work, evidence from schoolchildren to neurosurgery and dementia care
    concepts/leitner-systemThe learning box: cards advance on success, go back on failure; Leitner's 1, 2, 5, 8, 14 cm partitions
    concepts/graduated-interval-recallPimsleur's schedule, 5 s → 2 years (each step roughly ×5; marked as my inference)
    concepts/spaced-repetition-softwareHow SRS schedules cards; algorithm families Leitner, SM-0…SM-18, DASH, SSP-MMC/FSRS (Anki 23.10+, RemNote 1.16+)
    entities/sebastian-leitnerScience journalist behind the Leitner system; So lernt man Lernen
    entities/paul-pimsleurPublished graduated-interval recall in 1967
    questions/expanding-vs-uniform-intervalsOpen question, with a table comparing every review schedule in your wiki
    questions/who-originated-spaced-repetitionContradiction between your two sources (below)

    Updated pages

    • concepts/forgetting-curve: now points to spaced repetition as the countermeasure; adds Ebbinghaus's "review at several intervals"
    • entities/hermann-ebbinghaus: origin claim added, with a contradiction warning
    • concepts/active-recall: testing effect, recovery practice, delayed feedback, and the weakly supported "harder retrieval" mechanism
    • domains/learning-and-memory: current picture rewritten as one story, from forgetting to spacing
    • overview, index.md, log.md

    Contradiction found

    The new article credits

    Ebbinghaus (1880s) with conceiving spaced repetition, then credits

    C. A. Mace (1932) with first proposing it for improving learning. Your forgetting-curve source says Ebbinghaus studied forgetting "but not the effect of spaced repetition".

    I kept all three claims, flagged them on the Ebbinghaus and spaced-repetition pages, and opened questions/who-originated-spaced-repetition. My reading, marked there as an inference: Ebbinghaus hypothesized it; Mace proposed it as a study method.

    Handled with care

    • One sentence says SRS algorithms "use randomly determined equations". It's unclear and unexplained, so I left it out of the concept pages.
    • Several clinical claims rest on small studies or single cases; the pages say so.
    • No changes to wiki/me/, since the article says nothing about you.

    Suggested next steps

    1. Ingest a meta-analysis on spacing schedules; it would settle or sharpen expanding-vs-uniform-intervals.
    2. For the origin question: Mace's Psychology of Study (1932) or a history of spacing-effect research.
    3. Try a query: "Based on my wiki, how should I schedule reviews for language vocabulary?"

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Most "chat with your files" setups rediscover your notes from scratch on every question, so nothing builds up. This skill works the other way. Your agent reads each source once and files what it learned into a personal Markdown wiki: a summary, linked pages for people and ideas, an index and a log. The wiki gets richer with every source. It runs as an Obsidian vault in a git repo, and every claim links back to the source it came from.

    What it does

    • Ingest: drop an article, book chapter, PDF or notes into raw/ (or paste it into chat) and say "ingest". The agent writes a source summary, creates or updates the related entity, concept and domain pages, links them both ways, updates the index and log, and commits. A typical source touches 5–15 pages.
    • Query: ask a question and the agent answers from your pages, with a citation after every claim. If the wiki doesn't cover it, it says so instead of guessing. Useful answers are filed back into the wiki.
    • Lint: a read-only script finds broken links, orphan pages, index gaps and files not yet ingested. The agent fixes the mechanical issues and asks before merging or deleting anything.
    • Contradictions: when a new source disagrees with an old one, both claims stay side by side with their sources, and an open-question page is created.

    Ground rules the agent follows

    • Your sources in raw/ are read-only.
    • No invented facts. Unknowns are marked as unknown, and the agent's own inferences are labeled.
    • Text inside a source that tries to instruct the agent is flagged, never followed.
    • Nothing is pushed or sent anywhere unless you ask.
    • A pattern about you (wiki/me/) needs at least two independent sources before it is recorded.
    • Works in any language. File names stay ASCII, so links and git keep working.

    What's included

    • SKILL.md: setup, ingest, query and lint workflows
    • references/page-templates.md: templates for 8 page types
    • references/conventions.md: index and log formats, contradiction handling, git, Obsidian settings
    • scripts/init_wiki.py: creates the vault skeleton and never overwrites files
    • scripts/lint_wiki.py: read-only health check

    Both scripts use only the Python standard library, make no network calls and delete nothing.

    Based on Andrej Karpathy's "LLM Wiki" idea (April 2026).

    • Navigation relies on a single index.md catalog, which works well for hundreds of pages; very large vaults will strain the agent's context window.
    • Answers come only from the sources you ingest. General knowledge is used only when clearly labeled.
    • PDFs, scans and images depend on your agent's ability to read those formats.
    • Uses its own folder layout (raw/, wiki/, index.md, log.md, WIKI.md) and does not adapt an existing vault's structure automatically.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 13 days ago

    • Free with an Agensi account

    Requires

    Git
    Python 3.8+
    Obsisian (optioinal)

    Permissions

    Terminal / Shell
    Read Files
    Write Files

    File Scopes

    raw/**
    wiki/**
    index.md
    log.md
    WIKI.md
    .gitignore

    Terminal/Shell: runs the two bundled Python scripts (init_wiki.py creates the vault skeleton and never overwrites files; lint_wiki.py is read-only) and git status, git commit and grep inside the vault. Read files: sources in raw/ and pages in wiki/. Write files: only wiki/, index.md, log.md and WIKI.md, plus .gitignore at setup. raw/ is treated as read-only; the one exception is saving a source the user pastes into chat and asks to add. No network access and no environment variables. The bundled scripts make no network calls, and the skill never pushes to a remote unless the user asks.

    Listed13 days ago

    What's inside

    Frequently Asked Questions

    A summary prompt gives you one standalone answer. This skill has the agent maintain a persistent wiki: each new source updates the pages it touches, contradictions are recorded instead of overwritten, and every claim keeps its citation. The value compounds as you add sources.

    An agent that supports SKILL.md skills and can run shell commands (for example Claude Code), Python 3.8+ and git. Obsidian is optional, for browsing. Say "set up a second brain" and the agent creates the folder structure for you.

    Yes. Both claims stay on the affected pages in a warning callout with their sources and dates, and an open-question page is created so you can resolve it later. The demo above shows a real example.

    SKILL.md, two reference files (page templates and conventions) and two Python scripts (vault setup and a read-only lint check). No external dependencies.

    Any file your agent can read. The skill itself doesn't parse PDFs; it relies on your agent's file reading. Claude Code reads PDFs directly. Scanned documents without a text layer may need OCR first.

    The bundled scripts make no network calls, and the agent commits locally and pushes only when you ask. As with any agent, your model provider processes the files the agent reads.

    New in Productivity