Scheduled Peer Review

---
name: scheduled-peer-review
description: Set up and validate the opt-in GitHub Actions scheduler for deterministic Canvas group peer-review pairing.
---

# Scheduled peer-review pairing

This is an advanced-user deployment. It installs a course-owned GitHub Actions
workflow that periodically runs `peer_review_schedule.py`, which gates the
configured local-time window and delegates Canvas work to the existing,
idempotent `peer_review_assign.py` tool.

## Ownership boundary

- Canonical tools, examples, and this skill live in `canvas-toolbox`.
- The concrete workflow belongs in the course repository at
  `.github/workflows/peer-review-sync.yml`.
- The course schedule belongs at `.canvas/peer-review-schedule.yml`.
- Do not install either automatically through `cb_flatten` or `cb_update`.
- Canvas credentials belong in GitHub Actions secrets. The course ID may be a
  repository variable; never commit a token.

## Setup

1. Confirm the course repository is a flattened toolkit install and that
   `lib/tools/peer_review_assign.py` is present.
2. Copy `scaffold/peer-review-schedule.yml.example` to
   `.canvas/peer-review-schedule.yml` and set the assignment/group set, term
   window, lock time, and `America/Denver` (or the course's actual IANA zone).
3. Copy `scaffold/peer-review-sync.yml.example` to
   `.github/workflows/peer-review-sync.yml`.
4. Authenticate the GitHub CLI for the course repository, then use the
   secret-safe helper from the course repo:

   ```text
   uv run python lib/tools/github_actions_setup.py --repo OWNER/COURSE
   uv run python lib/tools/github_actions_setup.py --repo OWNER/COURSE --apply --yes
   ```

   The helper reads only `CANVAS_API_TOKEN`, `CANVAS_BASE_URL`, and
   `CANVAS_COURSE_ID` from `.env`; it writes the first two as repository
   secrets and the course ID as a repository variable. It never prints secret
   values. Without `--apply --yes`, it is a dry run.
5. Run the workflow manually with `apply` unchecked and verify the counts.
6. Validate against a real Canvas sandbox. Only then set `allow_enrolled: true`
   for an enrolled production course and enable the schedule.

## Safety properties

- The schedule is dry-run by default.
- The local lock time is evaluated with `zoneinfo`; the GitHub cron is only a
  polling cadence, so DST changes do not require editing the workflow.
- Runs outside the configured window or before the lock time are no-ops.
- Pairing writes remain idempotent and read-back verified through
  `peer_review_assign.py`.
- The scheduler prints counts/status only; it does not print student names or
  submission contents.
- A non-zero tool result fails the GitHub job and surfaces through GitHub's
  normal notification/history mechanisms.
- The setup helper never uses `gh secret set --env-file .env`; it allowlists
  the three expected keys so unrelated `.env` values cannot be uploaded.