Ask Your Corpus

    1

    Zero-dependency local retrieval for agents: search any folder of markdown/text files, rank matches (whole-phrase first), and answer with file citations. No embeddings, no vector database, no network, read-only.

    $3.99

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Ask Your Corpus

    Example session with this skill installed

    "What did my notes and postmortems say about why the DB migration failed?" → ranked sources with counts + a context.md the agent answers from, citing file paths.

    • Read your context and instructions
    • Compiled the ask your corpus

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

    About this skill

    Give your agent retrieval over any folder of .md/.txt files — the way it works on a real six-figure corpus — without installing embeddings, a vector database, or an MCP server. Two small pure-Python scripts, standard library only, strictly read-only, zero network.

    Long-context models (100K+) changed the economics of retrieval: you often do not need vectors. You need the RIGHT few files in the context window, cited.

    • Whole-phrase first: files containing your exact question score highest (occurrence count). That single decision answers most real questions.
    • Word-OR fallback: no phrase match? Rank by the sum of individual word hits — the graceful degradation that survives typos and multi-topic questions.
    • Citations by construction: the context block marks each source with a ### source: header; the closing instruction tells the model to cite every claim with that path. The agent answers from text, not from memory.
    • Budget-aware assembly: per-file limit for small files (include whole), match windows for big ones, and a global character cap so the block always fits the context window you have.

    Give context.md to your agent (Claude, GPT, llama.cpp, whatever you run). It answers with per-claim citations.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.

    Ranking is substring-based, not semantic (car does not find automobile). Reads text files only (.md/.txt default, extendable with --ext) — no PDFs. Very large corpora re-read files per query (fast up to ~100k files thanks to the OS disk cache).

    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 buying a skill.

    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 15 days ago

    Software
    Python 3.8+
    Folder of .md
    or.txt files

    Python uses the standard library only, so there is nothing to install. Any folder of files can be the corpus.

    Permissions

    Read Files
    Write Files

    File Scopes

    READ-ONLY:
    any
    corpus
    folder
    the
    user
    points
    at
    (-C/--corpus
    default
    current
    directory)
    WRITE:
    only
    single
    output
    file
    passes
    with
    --out
    (the
    assembled
    context
    block)

    Purpose: deterministic retrieval over a local text corpus with citations. corpus_search.py only reads files and ranks them by exact-phrase first, then word falls. build_context.py assembles a prompt-ready context block and optionally writes it to one user-chosen --out path. No outbound network at all, no command execution, no shell spawning, no access to environment variables or credentials. All file access is inside the user-selected corpus.

    Listed16 days ago
    Updated7 days ago

    What's inside

    Frequently Asked Questions

    No. The technique leans on long-context models: get the right few files into the window with occurrence-based ranking, then answer with citations. Real corpora of 6,500+ transcripts answered this way daily.

    Never. Both scripts are strictly local, read-only (plus one optional --out file). No outbound network connection exists in the code.

    corpus_search.py --json returns machine-readable hits; build_context.py produces a plain-markdown block ending with an instruction to answer only from the cited sources. Feed it to Claude, GPT, or llama.cpp — your stack decides.

    If ranking is not enough for your corpus, pair this with an embeddings pipeline — but most note/doc/transcript corpora start answering correctly with phrase-first ranking alone, at zero install cost.

    New in LLM Ops