Code assignments
Canvas-submittable code is a strong first-skill candidate even without a formal regression harness. The agent still finds the real spec, checks constraints, and runs whatever verification is available.
Four-step local setup
Download. Name your school. Log in. Choose a skill.
Copy one prompt. Your agent downloads Canvas Pilot, asks your school, opens Canvas login, then recommends the first durable skill.
1. Download locallyThe Agent clones or carefully updates the repository.
2. Tell it your schoolIt finds the official Canvas login instead of asking for config jargon.
3. Log in privatelyCredentials and 2FA stay in the Canvas browser.
4. Choose the first skillOpportunity analyzes recurring work and stops for your number.
School first
The first question is human: which school do you use Canvas through?
Login in the browser
The agent opens the official Canvas login and keeps secrets out of chat.
Opportunity runs next
It analyzes recurring work and asks which candidate should become the first skill.
First run
Four steps, in this order.
You answer one human question and complete one browser login. The Agent handles the local setup and first-skill analysis.
-
01
Download
The local Agent clones Canvas Pilot or updates an existing copy without overwriting local configuration.
-
02
Name your school
The Agent asks which school you use, then resolves the official Canvas login. It asks for a URL only when needed.
-
03
Log in
An interactive Canvas browser opens. You enter SSO and 2FA there, never in the Agent chat.
-
04
Run Opportunity
The Agent compares recurring work, recommends the best first skill, and stops for your numbered choice.
Know the fit before you install.
Canvas Pilot works best when the real instructions and required materials are reachable. It reports gaps instead of pretending a task finished.
Strong fit after course setup
Quantitative and business work
Math, accounting, economics, statistics, finance, and structured Word/Excel/PDF business work are strong candidates when the instructions and inputs are reachable.
Mostly objective Canvas quizzes
Supported choice, matching, true/false, and numeric quizzes fit well. Multiple attempts plus feedback visible before retry and keep-highest scoring make the correction loop especially useful.
Independent short-form work
Worksheets, short answers, and reading annotations can fit well when each response is brief and independently checkable, even when all responses together exceed 200 words.
A useful rule:
Judge the real task, its digital delivery path, and its feedback loop; do not rely on the file extension or total word count alone. If completion must happen inside an unfamiliar website UI, plan for a manual handoff.
Setup ends before coursework begins.
The pasted prompt configures the local tool and stops. Coursework starts only in a later, explicit request.
-
Set up
Install locally, complete browser login, assess recurring workflows, and stop for your choice.
-
Choose
Select one recommended candidate. Only then may Canvas Pilot verify and scaffold that course skill; selection does not authorize submission.
-
Scan
When you later say
scan canvas, the system writes a plan and stops again. -
Approve
You choose exactly which planned items may run. Unapproved work does not execute.
-
Execute
Approved workflows produce drafts, per-item results, and a report. Submission is never the silent default.
Review still matters.
A generated draft can be incomplete. A submitted status only means Canvas received an attempt; it does not prove correctness or a grade.
Private by design, explicit about limits.
Canvas Pilot runs on the machine that runs your agent. Login, course context, source files, drafts, and run state stay local.
You need
- Codex or a similar local agent with shell and file access.
- Git and Python 3.11 or newer.
- A desktop that can open an interactive Chromium window.
- Your normal Canvas login and 2FA method.
- Real assignment materials for any workflow you later approve.
Setup does not authorize
- Reading or executing pending assignments during setup.
- Uploading or submitting Canvas work.
- Publishing private course or credential data.
- Committing, pushing, or deploying from the local clone.
Keep secrets out of chat.
Enter passwords and 2FA only in the Canvas browser. Treat retained session cookies as sensitive local credentials.
Before you paste the prompt.
Can I use a browser-only chatbot?
No. Setup needs a local agent with shell access, file access, Git, Python, and permission to open an interactive browser.
Does setup start doing my assignments?
No. The setup prompt explicitly forbids scanning, planning, executing, uploading, and submitting coursework. The agent must verify setup and stop.
Can it work from an external course website?
It can read an accessible external specification or reference. It does not promise to operate an arbitrary third-party assignment portal.
What happens when a task is unsupported?
The workflow records an error or skip, explains the missing capability or material, and gives you a manual next step. It should never fabricate success.