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

Analyzing Data Files

A-81·Oct 28, 2026·Source: Web

Analyzing Data Files

Mandatory delegation workflow for any .xlsx, .xls, .csv, or similar file. Never read or compute on these files directly — all access goes through subagents.

Quick Start13 / 15

File uploaded → delegate to dataFile-context for a profile → summarize for the user in ≤8 lines → ask via question tool if they want to explore further → answer from the profile if possible, otherwise delegate again (and escalate to excel-reader only as last resort).

Recommendation▾

Examples are somewhat abstract (summarized outcomes rather than full concrete input/output transcripts) — add a fuller example showing the exact profile text returned and the exact final answer given to the user

Core Rules

Never read or calculate on the file yourself. No bash, no direct file reads. Only dataFile-context and excel-reader may touch the data.
Pass the exact file path as received — don't rewrite, infer, or shorten it.
Never invent figures. Every number given to the user must come from a subagent's response.
Retain the step 1 profile in conversation context — it's the source for direct answers later.
Limit re-delegation: max 1 re-delegation to dataFile-context and 1 escalation to excel-reader per user question.

Workflow15 / 15

Progress:

Step 1: Get general profile from dataFile-context
Step 2: Ask user if they want to explore further
Step 3: Answer from existing profile if possible
Step 4: Re-delegate or escalate for missing information
Final: Format response (result → details → caveats)

Step 1 — General Context (Always)

Delegate via task to dataFile-context:

Examine these files: <exact path(s)>.
Return the general profile: what the file is, structure by sheet, data quality, and key figures.
Do not modify anything. Be concise.

Multiple files → send together in one delegation.
If the user already asked a specific question, append User question: ... to answer it in the same round trip.
Summarize for the user in your own words, ≤8 lines, no long tables pasted.

Step 2 — Offer Further Exploration (Always)

Use the question tool once (verify its schema first — don't invent parameters):

Ask whether they want to explore further.
Up to 3 suggestions derived from the profile (e.g., "Sales totals by city", "Check duplicates in column X") plus "No, that's enough".
Allow free-text input.
Don't ask this in plain text if the question tool is available.

If they decline, close with a brief line and stop.

Step 3 — Answer From Context When Possible

If the request's answer literally appears in the step 1 profile (a figure, column name, date range, null count), answer immediately without delegating.

Counts as "not in context" (go to Step 4 instead):

Any new calculation (filter, group, sum, average) not already in the profile
Columns, sheets, or rows not detailed in the profile
Anything requiring verification or cross-referencing

When in doubt, go to Step 4. Never estimate or extrapolate from the profile.

Step 4 — Obtain Missing Information

4a. Re-delegate to dataFile-context with the specific question and known context so it doesn't repeat inspection:

File: <exact path>.
Known context: <relevant sheets/columns from the profile>.
Question: <exact user question>.
Do not repeat the general profile; perform only the necessary queries and return the figures.

Resume the prior subagent session if the task tool supports it.

4b. Escalate to excel-reader only if dataFile-context:

Declares it cannot obtain the information or states a limitation,
Fails or returns a tool error,
Returns incomplete/inconsistent results,
Or the question needs more than simple reading/aggregation (complex transforms, cross-sheet lookups).

File: <exact path>.
Question: <exact user question>.
Known context: <summarized profile>.
An attempt was made with dataFile-context and failed because: <reason>.
Answer only the question; do not repeat the general inspection.

If excel-reader also fails, tell the user honestly what couldn't be obtained and why. Never fabricate a result.

Recommendation▾

This skill is tightly coupled to specific subagents (dataFile-context, excel-reader) that may not exist in all environments — consider noting what to do if those subagents are unavailable

Follow-up Questions (Same File)

Skip steps 1–2. Go directly to step 3 or 4 using the existing profile. Only restart from step 1 if the file is new or changed.

Examples14 / 20

Example 1: Input: User uploads sales_2024.xlsx and asks "What's in this file?" Output: Delegate to dataFile-context for profile → summarize in ≤8 lines ("3 sheets: Sales, Returns, Summary. 12,450 rows, no major nulls. Key figures: total revenue $2.3M...") → ask via question tool with suggestions like "Revenue by region", "Top 10 products", "No, that's enough".

Example 2: Input: After profile, user asks "What's the total revenue for Q3 by region?" Output: Not literally in profile (requires grouping/filtering) → re-delegate to dataFile-context with known context and exact question → return the figures returned by the subagent, not estimated.

Example 3: Input: dataFile-context returns an error trying to cross-reference two sheets for a lookup. Output: Escalate to excel-reader with the failure reason, file path, question, and context. Report result or honest failure to user.

Recommendation▾

Quick Start section is a single dense line rather than a scannable step list; breaking it into 3-4 bullet steps would improve immediate scannability

Best Practices

One delegation per task, bundling all files together.
Include full context in delegation messages (path, question, known profile) to avoid wasted exploration steps.
Prefer dataFile-context first; excel-reader is the expensive last resort.
Request aggregates, not raw rows, unless the user specifically needs row-level data.
For large or multiple independent files, launch parallel task calls to dataFile-context.
Mention sensitive data (IDs, tax numbers, bank accounts, salaries) only if explicitly requested.
Respond in the user's language.

Final Response Format

Direct result (lead with the figure/answer).
Supporting details (brief table only if useful).
Assumptions and data quality warnings, only if relevant.

Common Pitfalls

Reading or computing on the file directly instead of delegating.
Rewriting or guessing the file path.
Presenting a number not returned by a subagent.
Re-delegating or escalating more than once per question.
Asking the exploration question as plain text when the question tool is available.
Pasting long raw tables instead of a concise summary.
Jumping straight to excel-reader without trying dataFile-context first.