Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Obsidian Ingest

    1

    Send your agent a YouTube URL and get back a few notes worth keeping, not another summary to sort through.

    $4.99

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Obsidian Ingest

    Obsidian Ingest

    Example session with this skill installed

    • Read your context and instructions
    • Compiled the obsidian ingest

    I wrote two drafts from the video "I replaced my entire stack with Postgres..." by The Coding Gopher (about 10 minutes). Both are in the Inbox's AI Drafts folder for your review:

    • 00 Inbox/AI Drafts/skip-locked-postgres-queue.md: FOR UPDATE SKIP LOCKED as a job queue in a plain Postgres table (linked to your PostgreSQL note).
    • 00 Inbox/AI Drafts/materialized-views-dashboards.md: materialized views with REFRESH MATERIALIZED VIEW CONCURRENTLY for dashboards instead of a warehouse (linked to your PostgreSQL note).

    Searches: "postgres postgresql", "skip locked queue", "jsonb gin index", "materialized view refresh". Only your existing Databases/PostgreSQL.md matched. It already covers the JSONB/GIN point, so I skipped that idea. The queue and materialized-view ideas had no existing notes, so they became new drafts.

    Not written: the video also covers full-text search with tsvector and pg_trgm, pgvector HNSW, PostGIS GiST, BRIN indexes for time series, and PostgREST/row-level security. I left these out to keep to the few ideas most worth rereading. The video's claims about throughput and the "replace everything" thesis are the speaker's, and I attributed them that way in the drafts.

    The first draft, as written to the vault

    
    # SKIP LOCKED turns a Postgres table into a job queue
    
    The Coding Gopher says a background worker queue built on a plain SQL table usually runs into contention: when two workers try to grab the same pending job row at the same time, one locks it and the other waits. Adding `FOR UPDATE SKIP LOCKED` to the select changes that. Each worker takes the first available row and locks it; if it hits a row another worker already locked, it skips that row and takes the next one instead of waiting.
    
    The speaker claims this makes a standard relational table a highly concurrent, wait-free queue that can handle thousands of jobs per second. Those throughput numbers are his claim, not a benchmark shown in the video. (02:51)
    
    ## Related Notes
    
    - [[Databases/PostgreSQL.md]]: Overview of the Postgres setup and indexing notes this queue pattern builds on.
    
    ## Source
    
    - Source: I replaced my entire stack with Postgres...
    - URL: https://www.youtube.com/watch?v=TdondBmyNXc
    - Creator: The Coding Gopher
    - Relevant timestamps: [02:51](https://www.youtube.com/watch?v=TdondBmyNXc&t=171s)
    

    I made three edits to the real output

    • Shortened the sandbox paths to vault-relative ones.
    • Dropped the agent's closing line about correcting a timestamp, along with the hand-corrected "(02:45…)" aside in the draft.
    • Removed one stray "Related:" sentence from the draft.

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

    About this skill

    Send your agent a YouTube URL and get back a few notes worth keeping, not another summary to sort through.

    What it does

    • Reads the whole source first. The agent reads the full English transcript in timestamped pages. Writing is refused until it has read all of it.
    • Checks what you already know. For each idea it searches your allowed notes and reads the closest matches before writing.
    • Writes a draft, an edit suggestion, or nothing. A new idea becomes a draft. An addition to an existing note becomes a separate, review-only suggestion. An idea you already cover, or a weak one, produces nothing.
    • Traces every idea to the source. Each note carries the video title, URL, creator and clickable timestamp links.

    Expect 0 to 3 notes per video. Zero is a valid result: let's-plays, vlogs and general career advice produce nothing.

    How it works

    The skill pairs these instructions with a small open-source Python runtime (a local MCP server and CLI) that your agent installs from GitHub. The runtime fetches captions, searches and reads your vault within the folders you allow, and writes files. Your agent decides what is worth keeping.

    Installation is agent-led: it asks for your vault path, allowed and excluded folders, shows a plan, waits for your approval, backs up host configuration and reports concrete undo steps.

    Where it stops

    • The writer is create-only. It never edits, overwrites, moves or deletes existing notes. Everything lands in one folder, 00 Inbox/AI Drafts/ by default.
    • You review every draft and apply every suggestion yourself.
    • It is an alpha release for Linux, tested on Claude Code, Codex, opencode and Hermes.

    Source and full documentation: https://github.com/YousefElGhoul/obsidian-ingest (Apache-2.0).

    YouTube only; Instagram and TikTok URLs are recognized but not extracted. Needs usable English captions: nothing is transcribed, and there is no OCR, speech-to-text or video inspection. Vault search is literal word matching, not semantic. Sponsor filtering is best-effort. Notes attribute claims to the source but do not verify them. Linux only; native Windows is out of scope. Alpha release.

    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

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 1 day ago

    Requires

    Youtube
    Linux
    Python 3.13+
    uv (Python tool runner)
    Obsidian vault
    Claude Code, Codex, opencode or Hermes
    Python
    Node.js
    Git

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Network Access

    Allowed Hosts

    youtube.com

    File Scopes

    00
    Inbox/AI
    Drafts/**
    **/*.md
    Listed1 day ago

    What's inside

    Frequently Asked Questions

    No. The writer is create-only and puts everything in one drafts folder. Changes to existing notes arrive as separate suggestion files that you apply yourself.

    That is by design. If the ideas are weak, already in your vault, or the video is a vlog or general advice, the skill writes nothing and says why.

    Only the folders you allow at install, minus the paths you exclude. On trusted setups the agent's other host tools are outside that boundary; Hermes offers a restricted mode without shell or file tools.

    Ask your agent to undo the install. It shows a plan, removes the MCP registration and skill link, and restores the host settings it backed up. Your vault and notes are left alone.

    New in Productivity