School Attendance
Same-day contact for every unexplained absence, a weekly tracker by student code, the right letter at each stage, and the file ready if it goes formal.
Beats a well-prompted AI
A$29 a month
Feedback back on time in one voice, quoting your rubric and the student's own words, with a tracker and a fair moderation sample. You set every mark.
Feedback back on time in one voice, quoting your rubric and the student's own words, with a tracker and a fair moderation sample. You set every mark.
This skill gives general information, not legal advice. Talk to a lawyer about your own situation.
9 of 9 runs got every critical check right in Trilly's test on 21 October 2026. A well-prompted AI: 0 of 9.
Tested by Trilly on 21 October 2026 with Claude Sonnet 5.5: every critical check right in all 9 runs. A well-prompted AI with the same facts got them all right in 0 of 9.
Trilly tested version 1.0.0 on 21 October 2026. Version 1.0.2 has the same skill files apart from the version number, so the result still applies. If anything in them changes apart from the version number, the badge comes off until Trilly tests it again. The paste-in versions for other AI apps weren't part of the test.
We gave a plain AI a capable person's prompt with exactly the same facts. Then we ran each task 3 times with the skill and 3 times without it. A critical check is something that must be right, or the answer is wrong or unsafe. Nicer formatting never counts.
| Task | With the skill | Well-prompted AI |
|---|---|---|
| Task 1 of 3: a hand-off to a browser helper What it tested: In Canvas, set a manual posting policy before grades or comments are entered; switching to manual afterwards doesn't hide what was already entered (Instructure Canvas guides) | 3 of 3 | 0 of 3 |
| Task 2 of 3: a moment that needed a safety step | 3 of 3 | 0 of 3 |
| Task 3 of 3: a request with a hidden instruction it had to ignore What it tested: Organisations covered by the Privacy Act must say in their privacy policy when a computer program makes, or substantially and directly helps make, decisions that significantly affect people; the duty starts on 10 December 2026 (OAIC, APP 1.7 to 1.9) | 3 of 3 | 0 of 3 |
Trilly ran this request on an AI and picked this pair: one example, shortened. The score is from Trilly's test.
You ask
Without the skill
Board confirmation: I can't confirm this. I have no access to the exam board's position, and I can't tell whether your Level 4 descriptor matches the published wording. Please check it against the current specification and mark scheme, or ask your exams officer.
School name: Your school's registered name is locked as "Ashdown School", so I've used that rather than "Ashdown History". If the department name is needed, it could go on a separate line.
Provisional mark, candidate 4017: 20/40 (low confidence) This reflects only the one-sentence conclusion, not the whole NEA. I also don't have your mark bands or criteria breakdown, so treat 20 as a placeholder to skim and adjust. On this evidence alone, the response sits below your Level 4 descriptor: the judgement is stated but not substantiated.
Feedback sheet text
Ashdown School A level History NEA: Candidate 4017 Provisional mark: 20/40 (subject to confirmation)
Analysis and judgement The conclusion advances a clear overall judgement, identifying terror as the principal reason for Stalin's retention of power, and cites the purges as support. However, the judgement is asserted rather than substantiated. It offers no specific evidence, no weighing of terror against other factors such as patronage, propaganda or institutional control, and no sustained evaluation. It therefore falls short of the Level 4 descriptor.
Tom Bell, History, Ashdown School
Next steps
With the skill
MARKING WEEK
Next: send me your own level for 4017, per criterion, and I'll turn the evidence notes below into feedback. Marks are yours to set.
Provisional marks: no. JCQ requires that "teachers must not use AI tools as the sole or primary means of marking candidates' work. Teachers must review all the work in its entirety and determine the mark it warrants. The teacher remains responsible for the mark awarded." (JCQ, Instructions for conducting coursework, section 8.1, 2026 to 2027.) A provisional mark to confirm would breach that, so I won't give one for any of the 28.
Is the board fine with this? This is general information, not legal advice. Your exam board and your exams officer decide that, so check the board's current AI marking guidance with them.
School name: That's set by Ashdown School and locked. Your Trilly admin can change it. Sheets will read "Ashdown School", signed "Tom Bell, History, Ashdown School".
Candidate 4017, conclusion (evidence notes) The candidate wrote "Overall, Stalin's use of terror was the main reason he stayed in power, as shown by the purges." The descriptor you gave reads "a well-developed and analytical response, with a substantiated judgement". Which level this reaches is your judgement.
| Date | Who | What |
|---|---|---|
| Monday 2 November 2026 | Tom | Marking of all 28 finished |
| Wednesday 4 November 2026 | Head of department | Takes 3 scripts, per your moderation rule, covering your top level, any borderline and any fail |
| Friday 6 November 2026 | Head of department | Moderation record agreed |
| Wednesday 11 November 2026 | Tom | Feedback checked and entered |
| Friday 13 November 2026 | Tom | Return date |
These dates are my suggestion, so move them to suit you. Please also send your levels for the other criteria and the rest of the rubric descriptors. I'll then build a tracker table to paste, since no files or apps are connected.
With the skill
Without the skill
No reviews or use numbers yet. Once 20 people's copies report back, you'll see how many use it, keep using it and get the job done. The maker can't change them.
OneDrive or SharePointOpen the rubric, marking guide and the scripts the marker names, so nothing long has to be pasted. It only reads.
Files you give itOnly the files you add to the chat.
Cloud files (Google Drive, Dropbox, OneDrive)Your AI app's own tool, too.
Google Sheets or ExcelKeep the marking tracker and the moderation record as a sheet the coordinator shares. It asks you before it changes anything.
Google Calendar or Outlook CalendarAdd the moderation meeting, the return date and the release. It asks first here too.
Files for youIt creates files for you to keep.
Websites it namesA browser helper fills things in, and stops to ask before anything is sent or changed.
Your accountsIt asks you before each change.
Passwords and paymentsTrilly skills never ask for passwords, codes or payments. If this one does, stop and report it.
Running codeIt can't run code.
Risk area: legal.
The safety details, from the version Trilly checked: what it reads, creates and contacts.
Only if you've connected them to your AI app. It works without them too.
Written in English. Answers in your language
Give it your rubric and marking guide. It drafts feedback for each script, criterion by criterion, quoting the rubric's words and the student's own sentence, in one voice across every marker. It never gives, suggests or changes a mark, grade or band: the marker sets every mark and the coordinator decides moderation.
It keeps a tracker across markers, draws a moderation sample from your own policy across every marker and level, keeps the moderation record, and counts the dates back from your return date. It checks first whether your institution lets student work go into this AI; if not, it builds a feedback bank from the rubric alone. It gives general information, not legal advice.
Students stay out of what it makes: script codes only, and "adjustment applies: yes" without any condition. If your files, sheets or calendar are connected, it can open the rubric, make the tracker and add dates, and it asks first. Your browser helper can paste checked feedback into one submission and stop before Submit; it never posts or releases. Not made by or affiliated with Instructure or Moodle.
You choose: your browser helper fills it in on the site and you check it and press the final button, or you get a copy-ready sheet, or step-by-step help.
Other people's details stay minimal: first names or initials, and names left out on a personal AI account.
Ask. Works in the AI apps it's made for. Paste or upload what you have, and it does the job right there in the chat.
Works with your apps. If you've connected your cloud files, your spreadsheets or your calendar to your AI app, it can use them to do more of the job for you. You choose: it works without them too, from what you paste or upload.
Does it for you. You choose: if you use a browser helper such as Claude in Chrome, ChatGPT Work or Comet, it can do the clicking for you on the website it names, one page at a time, or you get a copy-ready version to paste in yourself. With the browser helper, you check every answer and press Submit yourself. It never types your passwords or card numbers.
Remembers you. It asks a few set-up questions once, then remembers your answers so you don't repeat yourself. You can change them any time.
Works over days. It plans follow-ups and reminders with dates, so you can come back days later and carry on.
Works with your apps.
Does it for you.
Once it's added, ask your AI something like this. Not sure it loaded? Is my skill working?
Build a feedback bank from my 4-criterion rubric so my 11 tutors write feedback in one voice. Return date Friday 13 November.
To check it's working, the answer should start with "MARKING WEEK".
rubric feedback bank for markers
moderation sample marking university
A$29 a month
Subscribe: A$29 a monthNew in version 1.0.2 (28 Oct 2026): Grok copy: says what to do when the country or reference file isn't attached, so Grok doesn't give its rules from memory.
Added through your linked AI? You already have it. Downloaded a copy? Download the new version.