Section 01 · Failed builds
When a build fails.
There are three distinct reasons a build can fail. They look similar from the outside — the job stopped, something went wrong — but the cause is different each time, and so is what happens next. Here is the full picture for each one.
Conversion failed (pipeline error)
A pipeline error means something broke inside our extract, plan, synthesise, QA, or packaging process — a server fault, a timeout on our end, or a step that returned an unexpected result. It has nothing to do with your file or your content.
An error occurred in one of our six pipeline stages (Extract, Plan, Synthesise, Auto-QA, Package, or Deliver) that caused the job to stop. This is a system fault on our side, not an issue with your document.
Full automatic credit refund. The credits spent on that job are returned to your balance immediately. No email required, no form to fill out. The refund appears in your Credit history within minutes of the failure being detected.
Check your Credit history to confirm the refund has posted (it always does, within minutes). Then try the same file again — most pipeline errors are transient and the retry succeeds. If the same file fails twice in a row, email us with the job ID and we will investigate.
QA below threshold (citation-tracing failed)
After building your skill, our Auto-QA stage samples claims from the result and traces each one back to a real passage in your source document. If too many claims cannot be verified — meaning the skill asserted something the source doesn't actually say — we stop rather than ship a skill you can't trust.
This is not the same as the skill being wrong. It usually means the model made reasonable inferences that went slightly beyond what the source text can prove. The claim might be true, but we can't point to a sentence that backs it up. Our bar is strict because that is the whole point of the product: everything in your skill should trace back to something real you own.
When QA falls below our 95% verification threshold, you see the QA results screen. That screen shows you exactly which claims failed, why each one failed (no source match vs. partial match vs. date mismatch), and the overall pass rate. You then choose one of three paths:
Auto-QA checked a sample of claims from your skill against your source document. The percentage of claims that could be verified fell below our 95% threshold. The skill is otherwise complete — QA is the final check before delivery, not an early-stage failure.
We pause delivery and show you the QA results: total claims checked, how many were verified, the specific claims that failed, and why each failed. Nothing is shipped until you decide.
You have three options from the QA results screen:
- Re-run with the Premium model (Pro). The Pro model uses a larger context window and stricter grounding. It costs additional credits but typically clears the QA bar. Recommended if you need a clean, fully-verifiable skill.
- Accept the skill as-is. The unverified claims are flagged inside your skill so you know exactly where the gaps are. You still get a usable skill — you are simply aware of its limits. Some users do this when the unverified claims are minor inferences they are comfortable with.
- Full credit refund. If you would rather not keep a skill with gaps, you can request a full refund and the credits come back to your balance immediately.
Source unreadable (zero usable text)
The pipeline fails at stage 01 (Extract) when we cannot get any readable text from the file. This typically happens with three file types: scanned PDFs that are just images with no text layer, encrypted PDFs that block text extraction, and corrupt files where the data was damaged in transit or storage.
There is no point continuing to later stages if we cannot see words. A skill built from an empty extraction would be entirely hallucinated — which is the opposite of what SkillPill is for.
Our extraction stage read your file but found zero usable text characters. This indicates one of: a scanned image-only PDF (no embedded text layer), a PDF with password protection or DRM that blocks text extraction, or a file that arrived corrupt.
Automatic full credit refund. Credits return to your balance immediately. The job is closed and your source file is queued for deletion as normal.
Fix the file first, then re-upload. The specific fix depends on the cause:
- Scanned PDF (image-only): run OCR first. Adobe Acrobat, Apple Preview (Enhance Scan), or the free command-line tool
ocrmypdfall work. After OCR, the PDF has a real text layer and our extractor can read it. - Encrypted or DRM-protected PDF: remove the restriction. If it is your own document you encrypted yourself, open it in Acrobat or your PDF tool, remove the password, and re-save before uploading. If it is a purchased ebook with DRM, check whether the publisher offers a DRM-free EPUB — many do.
- Corrupt file: try re-downloading from your original source, or export a fresh copy from the source application (Word, Notion, Google Docs, etc.).
For more detail on which file formats work best, see Preparing your document.
Files deleted within 24h, automatically. Your source file is removed from our servers automatically — on delivery for successful jobs, and within 24h for failed jobs. We never retain your content beyond the build window.
Section 02 · Wait issues
It has been more than 24 hours.
Standard turnaround is up to 24 hours. Most jobs complete well inside that window, but pipeline congestion or unusually large documents can push a job toward the full 24h. If your job is approaching or has passed 24h, here is what is happening and what to do.
Check the status banner in the app
Open the app at app.skillpill.ai and go to your processing job. The status banner shows the current pipeline stage and elapsed time. If the job is still actively running — you can see it on a stage and the elapsed timer is counting — it has not timed out. It is still building.
If the status shows a failure or timeout state, the automatic refund will have already been issued and you will see the credits returned in your Credit history.
Between 24h and 48h: email us with your job ID
If your job has been running for more than 24 hours and you have not received a delivery or failure email, contact us at [email protected]. Include:
- Your account email address
- The job ID (visible on the processing page, format similar to
K7-4FC2-A91) - The file name or skill name you submitted
We respond within one business day and will manually check the job status.
Pipeline timeout — automatic refund
Any job that has not completed within 48 hours is treated as a pipeline timeout and receives a full automatic credit refund, per our refund policy. You do not need to contact us for this — the system handles it. The refund appears in your Credit history and you will receive an email notification.
After a timeout refund, you are free to re-submit. Timeouts are rare and typically caused by transient infrastructure issues; a retry almost always succeeds.
Priority turnaround on Plus and Premium. If you are on Plus or Premium, your jobs enter a priority queue. Large documents on Standard tier are the most likely to approach the 24h mark. Upgrade mid-cycle and we credit the difference.
Section 03 · Estimate surprises
The estimate was higher than I expected.
The credit estimate shown at the confirm screen can sometimes surprise people. The estimate is always shown before any charge — you approve it explicitly and can cancel before spending anything. But if the number seems higher than you expected based on the raw file, here are the three reasons that can push it up.
No surprises after confirmation. Cost is shown before any credits are charged. You see the number, you click Confirm, and only then does the job begin. If the estimate looks wrong to you, cancel at that screen — no credits are spent.
| Cause | Why it happens | How to reduce it |
|---|---|---|
| Language multiplier | Non-English documents require additional processing to extract, plan and synthesise accurately. Languages with complex morphology or CJK character sets cost more per page than plain English. | No workaround — this is inherent to the language. If cost is a concern, consider whether an official English translation of your document exists. |
| Page count after extraction | The estimate is based on the extracted page count, not the file's reported page count. Figures, tables, footnotes, bibliography, and appendices all count as pages. A 200-page business book may extract to 250+ pages once appendices and endnotes are included. | If you only need specific chapters, consider extracting them into a separate document first and uploading that subset instead of the full file. |
| Pro / Real-time toggles | Selecting the Pro quality model uses 2× the credits of Standard for the same document. Selecting Real-time delivery uses 4× credits on top of the base cost. Both toggles are applied at the confirm screen. | Switch quality to Standard and delivery to Standard on the confirm screen. The estimate updates live as you change these options. Standard quality passes QA for the large majority of well-structured documents. |
The estimate uses the formula: base page cost × language multiplier × quality multiplier × delivery multiplier. The base rate is approximately 100 credits for a standard 300-page English document at Standard quality. Shorter, clean, English documents with no real-time or Pro toggle can cost as little as 15 credits.
The estimate is conservative — it rounds up to avoid under-charging mid-job. The actual cost at delivery may be slightly lower if the document extracts more compactly than expected.
Section 04 · Out of credits
Not enough credits to run this job.
When you try to start a build and your credit balance is lower than the estimate, the job is blocked and you see the out-of-credits screen. The screen tells you the exact shortfall and gives you three options for getting back on track. Nothing has been charged yet at this point.
Top up — buy exactly what you need
The out-of-credits screen calculates the exact shortfall and offers a top-up pack that covers this job with a small surplus. Top-up packs are one-time purchases: no subscription, no commitment. The credits are added to your balance immediately and the job can start as soon as the purchase completes.
Top-up credits behave differently from monthly plan credits: they do not expire at the end of your billing cycle. They stay on your account until you use them or close the account. For full details on how top-up credits interact with monthly credits, see the refund policy — top-up packs.
Upgrade — more credits per month at a lower per-skill rate
If you are regularly running out of credits, upgrading your plan is cheaper than repeated top-ups. When you upgrade mid-cycle, we prorate the charge: you pay only for the remaining days in your current billing period at the new plan rate. The upgrade takes effect immediately and your credit balance jumps to the higher plan allowance.
| Plan | Credits / month | Approx skills / month | Price |
|---|---|---|---|
| Starter | 300 | ~3 | $29 / mo |
| Plus | 1,000 | ~10 | $83 / mo |
| Premium | 2,400 | ~24 | $179 / mo |
Wait for your cycle to refresh
Monthly credits refresh on your billing date. If you are close to that date and the job can wait, you may prefer not to spend on a top-up. The out-of-credits screen shows your next refresh date so you can judge whether waiting makes sense. Your job is saved and will be ready to submit the moment the credits arrive.
No build gets blocked mid-flight. Credits are reserved at the start of a job, not charged incrementally as it runs. If you have enough credits to start, the job always runs to completion — you will never have a job cut off partway through because credits ran out mid-build.
Section 05 · Install problems
The skill installed but isn't using my content.
The most common install issue is not a broken install — it is a skill that appears to be answering generically when it should be drawing on your specific content. There is a reliable ritual for diagnosing this in under a minute.
The verify ritual — "What is in your skill pack?"
Before any other troubleshooting step, ask your AI exactly this question:
A working install responds with your specific chapter titles, frameworks, and named concepts from your document. A failing install (or no skill loaded) responds with something generic like "I'm trained on a wide variety of data" or lists stock topics.
If the answer is specific to your content, your skill is loaded and working. The issue may be with the question you are asking — make sure you are asking about something that is genuinely in your document.
If the answer is generic, work through the platform-specific checks below.
Checks by platform
- Claude Code: skill not responding to /[slug]
- Check that the skill folder is at exactly
~/.claude/skills/[skill-slug]/(macOS/Linux) orC:\Users\[you]\.claude\skills\[skill-slug]\(Windows). The folder should not be nested an extra level. Restart Claude Code after moving the folder — it discovers skills on startup, not live. Also confirm thatSKILL.mdis at the root of that folder, not inside a subfolder. - Claude Code: answers are generic even after typing /[slug]
- The skill loads fresh at the start of each conversation. Type
/[skill-slug]as your first message in a new conversation, not mid-thread. If you open an existing thread and then load the skill, earlier context may override it. Start fresh. - ChatGPT: still giving generic answers after upload
- Check the Knowledge section in GPT Builder — all files should appear as blue chips. If any are missing, re-upload them individually. Large packs can take several minutes to index after upload; try the verify question again after waiting a few minutes.
- ChatGPT: using free tier, no GPT Builder
- GPT Builder (Custom GPT creation) requires ChatGPT Plus, Team, or Enterprise. On a free account, the alternative is to start a ChatGPT Project and upload the skill files to the project. Projects on the free tier do not persist context as precisely as a Custom GPT, but the skill files will be available within that project's conversations. See the full install guide for the Projects path.
- Claude.ai Projects: answers still generic after upload
- Large packs can take a minute or two to index fully in a Claude.ai Project. After uploading, wait until all files show as processed (look for a check mark or "indexed" status in the project's knowledge section), then start a new conversation inside the project — not a conversation that was open before the upload.
- Gemini Gems: skill not working
- In Gemini, both the knowledge files and the instructions field must be filled before saving. The instructions tell Gemini to prioritise your files. Without them, Gemini treats the uploaded files as supplementary context only. Re-open the Gem editor, confirm the instructions field is populated with the text from your pack, save again, and start a fresh chat.
- Gemini: no Gems option visible
- Gems require Gemini Advanced (Google One AI Premium, $19.99/mo). Without it, you can still upload skill files directly to a standard Gemini conversation for a single session. The files won't persist across sessions, but you can re-upload at the start of each conversation you need the skill for.
For the complete step-by-step install walkthrough for each platform, see the Install guide.
Section 06 · Account & billing
Account and billing questions.
Subscriptions, payment issues, email changes, and account deletion — each one addressed directly.
How to cancel your subscription
One screen, one click, no retention call. Go to Billing in the app sidebar and click Cancel subscription. Your access continues until the end of the current paid period — cancellation stops the next charge, it does not cut off access immediately.
Skills you built do not expire when you cancel. They remain in your account and are yours to download at any time. The SKILL.md files work independently of your subscription.
If you are on the 7-day trial, cancel before day 8 to avoid the auto-conversion to the Starter plan. One click from Billing — the trial ends, no charge follows, and you keep whatever was built.
Card declined or payment failed
When a renewal payment fails, you receive an email from us titled something like "Your payment didn't go through." Your account enters a short grace period — access is not cut off immediately, giving you time to update your card.
To resolve it: go to Billing → Payment method and update your card details. Once the card is updated, the failed payment is retried automatically. If the retry succeeds, your subscription continues without interruption and the failed invoice clears.
If you do not resolve the payment during the grace period, the subscription moves to a paused state. Your skills and account data are preserved; you just cannot start new builds until billing is resolved. Re-enter valid payment details to reactivate.
Changing your account email address
Your login email is managed through your account settings. Go to Settings → Profile and update your email address. You will receive a verification email at the new address — the change takes effect once you click the verification link.
If you are having trouble accessing Settings or the verification email is not arriving, email us at [email protected] from your current account email and we will handle it manually.
Deleting your account
Before deleting your account, download your skills first. Deletion removes all of your built skills from our servers permanently. Once the deletion is complete, the files are gone and cannot be recovered.
What account deletion removes:
- All built skills — download them before you delete if you want to keep them
- Your credit balance — including any unspent top-up credits (non-refundable after 14 days from purchase)
- Your billing history and invoices — download invoices from Billing before deleting if you need them
- Your account data and preferences
To delete your account, go to Settings → Account → Delete account. If you cannot find this or prefer to request deletion by email, contact [email protected] from your account email.
Subscription cancellation happens automatically as part of account deletion. You do not need to cancel separately first.
Download your skill files before deleting. Your SKILL.md files work independently in any AI that reads SKILL.md — Claude, ChatGPT, Gemini. Once they are downloaded and installed, you do not need a SkillPill account to keep using them.
Section 07 · Still stuck
Still stuck? Contact us.
If the above has not resolved your issue, email us. A founder reads every support email — no ticket system, no bot, no routing queue.
[email protected]
We respond within one business day, usually same-day on weekdays. To help us resolve your issue quickly, include the following in your email:
| Include | Why it helps |
|---|---|
| Your account email address | Lets us find your account without asking you to reply again |
| Job ID (format: K7-4FC2-A91, visible on the processing page) | Points us directly to the specific build without guessing |
| File type (PDF, EPUB, DOCX, etc.) | Narrows down the likely cause for extraction or format issues |
| What you expected vs. what happened | One or two sentences is enough — we read carefully and ask follow-up questions if needed |
For build failures that already issued an automatic refund, there is nothing you need to do on the billing side — just tell us what happened so we can investigate. For billing disputes or refund requests that fall outside the automatic cases, the same email works — the refund policy covers the full set of cases.