An AI skill is a structured system prompt or set of custom instructions that gives AI assistants like ChatGPT, Claude, and Gemini specialized capabilities. It includes a clear role definition, specific knowledge areas, step-by-step processes, and quality criteria that help AI perform specific professional tasks.

How does Skill This work?

Simply describe your professional expertise in plain text. Our free AI skill generator analyzes your input and transforms it into a structured system prompt. The skill is then graded on clarity, specificity, and usefulness. You can share your skill via a unique URL or copy it to use as custom instructions with any AI assistant like ChatGPT, Claude, or Gemini.

Is Skill This free to use?

Yes, Skill This is completely free. No account or payment is required. You can create unlimited skills and share them with anyone.

How is my skill graded?

Skills are graded on a scale from A+ to F (0-100 points). The grade reflects how clear, specific, and actionable your skill is. Higher grades indicate skills that AI assistants can use more effectively.

How do I use my skill with an AI assistant?

Copy your generated skill and paste it into the system prompt or custom instructions of any AI assistant. For ChatGPT, use Personalization > Custom Instructions. For Claude, add it to a Project's instructions. For Cursor or other IDEs, save it as a .cursorrules or CLAUDE.md file. The AI will then adopt that skill and perform tasks according to your expertise.

AI Skill Report Card

Building Compiled Second Brains

A-87·Sep 13, 2026·Source: Extension-page

Quick Start14 / 15

The core architectural shift: stop storing, start compiling.

vault/
  raw/        ← input buffer only, never the answer
  wiki/       ← compiled, cross-linked knowledge pages
  output/     ← built from wiki, not from raw or memory
  CLAUDE.md   ← compiled profile of the person, read every session

Minimal first action for a new vault:

Create the four items above.
Run the CLAUDE.md interview prompt (below) to seed the profile.
Drop one real source into raw/ and run the ingestion prompt.
Confirm the model wrote wiki pages, linked them to nothing yet (first source), and gave a three-sentence summary.

Recommendation▾

Add a bad-outcome example (e.g., a wiki that regressed into a filing cabinet) to contrast with the good examples for stronger before/after learning.

Workflow14 / 15

Progress:

Step 1: Set up the four-part structure (raw/, wiki/, output/, CLAUDE.md)
Step 2: Build CLAUDE.md via structured interview
Step 3: Ingest first sources one at a time, reviewing wiki pages as they're written
Step 4: Once 10-15 sources exist, set up the daily compilation loop
Step 5: At 50-100 sources, evaluate whether non-obvious connections are surfacing
Step 6: For active work, spin up project folders (Inputs/Process/Outputs/Feedback) instead of dumping into the general wiki

Step 1 — Structure. Nothing in raw/ is ever treated as an answer to a question — it's unprocessed input waiting for compilation. wiki/ is the only place knowledge lives. output/ (documents, posts, decisions) must be built from wiki pages, never from raw sources or from the model's unaided memory.

Step 2 — CLAUDE.md. This is not a static prompt. It's a living profile the model updates and reads before every session. Interview one question at a time; don't batch questions. Cover: identity/role, current goals, communication preferences, strengths/weaknesses, active projects. Write the result with clear headers at the vault root.

Step 3 — Ingestion. Each source should touch 10-15 wiki pages: new concept pages get created, existing pages get updated, links form to related pages, and contradictions with prior entries get flagged explicitly rather than silently overwritten. Always end ingestion with a short human-readable summary of what changed.

Step 4 — Daily loop. Automate (via Claude Desktop scheduled tasks) a daily pass that: files new raw items into the right wiki locations, flags stale notes (no update >2 weeks), checks for contradictions introduced by recent additions, and produces a morning brief of changes/links/flags/action items.

Step 5 — Density threshold. Before ~50-100 well-compiled sources, the wiki behaves like a slightly-better search index — this is expected and not a failure. The payoff (non-obvious cross-links surfacing on their own) appears after that density is reached.

Step 6 — Project isolation. For a bounded piece of work, don't mix it into the general wiki. Create a dedicated folder with Inputs/Process/Outputs/Feedback and its own CLAUDE.md defining the single goal, definition of done, and the model's specific role.

Recommendation▾

Include a concrete CLAUDE.md template snippet rather than just describing its headers, to make Step 2 more directly actionable.

Examples17 / 20

Example 1: New source ingestion Input: A user drops raw/transformer-scaling-laws.pdf into the vault and says "process this." Output: Model creates/updates wiki pages like wiki/scaling-laws.md, wiki/compute-optimal-training.md, links each to existing pages (e.g., wiki/chinchilla-paper.md), flags that this contradicts an earlier note claiming "bigger models always win," and returns: "Added 2 new pages, updated 3 existing ones, flagged one contradiction with your Jan note on model size — see wiki/chinchilla-paper.md."

Example 2: CLAUDE.md interview Input: User says "set up my second brain." Output: Model asks one question at a time — role, yearly goals, communication style, strengths/weaknesses, current projects — waits for each answer, then writes a structured CLAUDE.md with headers like ## Who I Am, ## Current Goals, ## Communication Preferences, ## Active Projects.

Example 3: Daily brief Input: Scheduled daily task runs. Output: "Filed 2 new notes from Inputs/ into wiki/marketing-strategy.md. Flagged wiki/q3-roadmap.md as stale (18 days, no update). No contradictions found. Recommend reviewing the new competitor-pricing note — it links to 3 existing pages and may change your Q3 assumptions."

Recommendation▾

Clarify what 'contradiction flagging' output looks like in wiki markdown (e.g., an inline flag syntax) so ingestion behavior is fully reproducible.

Best Practices

Treat raw/ as strictly disposable input, never a reference source for output.
Never let the model build output/ directly from raw material — force it through the wiki compilation step first.
Interview for CLAUDE.md interactively, not as a single giant prompt — quality drops when questions are batched.
Curate sources deliberately. Since compilation propagates into many linked pages, source quality matters far more than in a simple search/retrieval setup.
Expect and tolerate a slow start (first few weeks) — don't judge the system before ~50 sources are compiled.
Use separate project folders (with their own CLAUDE.md) for active, goal-bound work instead of polluting the general wiki.
Require the daily loop to always end in a human-readable brief — the human must stay inside the loop, not just receive silent automation.
This setup assumes Claude Desktop with file system access and scheduled tasks on a paid plan — don't attempt on free tier.

Common Pitfalls

Treating raw/ as searchable knowledge. If someone answers a question by scanning raw files instead of the compiled wiki, the system has regressed into a filing cabinet.
Skipping the CLAUDE.md interview and hand-writing a static prompt. This defeats the "living document" property — it should evolve and be re-read every session.
Expecting non-obvious connections before reaching source density (~50-100). Below that threshold, a search engine does the same job — don't oversell early results.
Ingesting low-quality sources without scrutiny. A bad source in a compiler touches many pages before it's noticed, unlike a bad document sitting alone in a library.
Mixing active project work into the general wiki. Bounded projects need isolated Inputs/Process/Outputs/Feedback folders with their own goal-specific CLAUDE.md.
Running this on free-tier Claude. Scheduled tasks and file system access are required for the compilation loop; without them the system is manual and will rot like any other note archive.