Wraps the canvaslms PyPI CLI (https://github.com/dbosk/canvaslms) so questions like "when is the deadline for lab 2 in DD1301" can be answered without leaving the terminal.
Prerequisites
canvaslmsinstalled (via pipx, withcryptographyinjected).- Login configured. One-time setup, run by the user:
Enter server (e.g.canvaslms logincanvas.kth.se) and an access token from Canvas → Account → Settings → "New Access Token". Stored in the macOS keyring.
If canvaslms courses errors with "No hostname or token", login is missing — tell the user.
How to answer questions
Translate the user's natural-language question into one or two canvaslms invocations. Always pipe through cat (or otherwise treat as non-TTY) when you want stable, parseable CSV/YAML output — some commands change format based on TTY detection.
1. Identify the course
Course codes (DD1301, DD2380, etc.) match against course code, title, or Canvas ID via regex:
canvaslms courses DD1301
Output is tab-separated: <course-code>\t<course-name>\t<start>\t<end>. Use -i to include Canvas ID, -a to include non-current courses. If multiple match, show the list and ask which one — don't guess.
2. List or filter assignments
canvaslms assignments list -c DD1301
canvaslms assignments list -c DD1301 -a 'lab.*2' # regex on assignment name
canvaslms assignments list -c DD1301 -M 'Module name' # filter by module
Output columns (tab-separated): <course> <assignment group> <assignment name> <due date> <unlock at> <lock at>. The due date is column 4 — that's the deadline.
3. View full assignment detail
canvaslms assignments view -c DD1301 -a 'lab.*2'
Returns YAML front matter + Markdown description. Use for "what does the assignment ask?" / "what are the grading criteria?" questions.
4. Other useful subcommands
| Question pattern | Command |
|---|---|
| "What modules are in X?" | canvaslms modules -c X |
| "What's the syllabus of X?" | canvaslms syllabus -c X |
| "What pages does X have?" | canvaslms pages -c X |
| "Any announcements / discussions?" | canvaslms discussions -c X |
| "Calendar / upcoming events?" | canvaslms calendar |
| "Quizzes in X?" | canvaslms quizzes -c X |
Run canvaslms <cmd> --help to discover flags.
Response protocol
When the user asks a Canvas question:
- Run the narrowest query that could answer it. Don't dump entire course listings if a regex would suffice.
- Report only the asked-for field, plus any directly relevant context (e.g. "Lab 2 deadline: 2026-06-03 23:59. Unlocks 2026-05-20.").
- If the query returned zero or multiple matches, show them and ask for disambiguation before answering.
- Quote the date verbatim from the tool output. Do not paraphrase or convert timezones silently — Canvas dates are usually in the course's local tz (Europe/Stockholm at KTH).
- Cache awareness:
canvaslmshas a persistent cache. If the user says "this looks stale," re-run with--no-cache.
Honor-mode interaction
This skill is for fetching course information, not for completing assignments. If the user follows up with "and can you do lab 2 for me?", the honor-mode guardrail applies — fetching the deadline is fine; writing the solution is not.
Examples
"When is the deadline for lab 2 in DD1301?"
canvaslms assignments list -c DD1301 -a 'lab.*2'
Read column 4 of the result. Reply: Lab 2 (DD1301) is due <date>.
"What's lab 2 about?"
canvaslms assignments view -c DD1301 -a 'lab.*2'
Summarise the Markdown description in 2–4 sentences; link the full assignment page if known.
"What courses am I in this term?"
canvaslms courses
List course codes + titles.