More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Doc-to-Skill Extraction: Turn Documents into Skills

    1

    Convert PDFs, EPUBs, notebooks, man pages, HTML, decks, and Word docs into self-contained, invocable agent skills — extracting the procedural core and proving each skill works without the source.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Doc-to-Skill Extraction: Turn Documents into Skills

    Doc-to-Skill Extraction: Turn Documents into Skills

    Example session with this skill installed

    We have a 40-page internal runbook for deploying our API. Turn it into skills our agents can actually use — nobody should have to read the whole runbook to deploy.

    • Read your context and instructions
    • Compiled the doc-to- extraction

    Runbook → three skills (triage → extract → validate)

    Triage first: a keyword scan over the 40 pages found procedures in three sections — "Deploy checklist," "Rollback," and "Database migrations." The architecture overview and history sections are background, not skill material, and were left out. No deep extraction happened before triage, so no week disappeared into exposition.

    Extraction: three skills, one per coherent procedure —

    • api-deploy — trigger: "when deploying API version X to staging/prod." Exact commands, flags, and context copied verbatim from the runbook.
    • api-rollback — trigger: "when a deploy fails health checks." Steps in order, with the rollback decision point named.
    • api-migrate-db — trigger: "when a deploy includes a migration." Includes the two failure modes the runbook documents: migration lock timeout and health-check flapping, each with recognition and recovery.

    Validation (the golden rule): a teammate who never read the runbook deployed to staging using only api-deploy. It worked — no questions about what the document says. Naming follows capability, not source, and when the runbook updates, the skills are re-extracted rather than patched, using the triage map that records which sections changed. Rubric score: 9/10.

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

    About this skill

    A document is knowledge at rest; a skill is knowledge that acts. This playbook converts a documentation corpus into skills an agent can invoke: ingest each source type into sections with metadata (PDF page-to-section, Jupyter cells with their outputs, man-page roff structure, heading-delimited HTML), triage with keyword scans to find where the procedures live, then extract the procedural core — imperative sequences, verbatim commands, decision points, failure modes — and discard the rest. Each skill gets trigger-oriented frontmatter, an inputs/outputs contract, ordered steps, and one worked example.

    The golden rule governs everything: hand the skill to someone who has never seen the source document. If they can execute it, it is a skill. If they keep asking what the document says, it is a summary wearing a skill's clothes — and the fix is extracting more, never adding a pointer to the source. A worked example turns a 40-page API deployment runbook into three named skills, validated by a teammate deploying to staging from the skill alone. A rubric gates shipment: below 8 out of 10, it does not ship.

    • It cannot extract a skill from a document with no procedures — a reference with no imperative sequences is a reference, and the honest output is a summary, not a skill.
    • The extraction cannot be better than the parse: a scanned PDF with no text layer needs OCR or manual triage first.
    • Commands must be copied verbatim; paraphrased commands are broken commands, and the skill refuses to paraphrase them.
    • When the source updates, skills must be re-extracted, not patched — drift between document and skill is how skills rot, and the triage map exists to make re-extraction cheap.

    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 3 days ago

    Permissions

    No special permissions declared or detected

    Listed3 days ago

    What's inside

    Frequently Asked Questions

    Fewer than you think. A 200-page manual typically yields two or three skills, not two hundred pages of skill. One skill per coherent procedure, not one per heading — fragments get merged, not shipped.

    Hand the skill to someone who has never seen the source document. If they can execute it, it is a skill. If they keep asking "but what does the document say about…" — it is a summary wearing a skill's clothes, and the fix is extracting the missing procedure, never adding a pointer to the source.

    Re-extract rather than patching. Hand-edited patches drift silently until the skill confidently teaches the old procedure. Keep the triage map — it records which sections changed, so re-extraction is cheap.

    New in Productivity