NOTE: Ask Claude Cowork to setup this scheduled task for you - all you need to do is add
*.github.comto allowed domains, Claude can follow the steps below on its own. It'll ask you for everything it needs to setup the task.
A Claude Cowork based scheduled task that produces a daily GitHub Radar report — a view of issues, PRs, reviews, mentions, and threads waiting on you across every org you belong to. The output is rendered into a self-contained HTML artifact (github-radar).
-
macOS with Claude Cowork installed and the Scheduled Tasks feature enabled.
-
A GitHub Personal Access Token (classic or fine-grained) with read access to the orgs and repos you want surfaced. Pick whichever style you prefer:
Classic PAT — tick these scopes:
repo(full control of private repos; needed for issue/PR reads in private repos)read:org(list orgs you belong to)read:discussion(read repo discussions)
Fine-grained PAT — set "Resource owner" to org/user for which you want to see private repositories. All other orgs/user will only list public repositories. Under "Repository permissions" set:
Contents→ ReadIssues→ ReadPull requests→ ReadDiscussions→ ReadMetadata→ Read (auto-selected, required)
And under "Organization permissions":
Members→ Read (so the script can enumerate orgs and detect solo-member scopes)
-
Create an empty scheduled task in Cowork. Open Claude Cowork → Scheduled Tasks → New scheduled task. Name it
github-radar(the name becomes the directory name, and the artifact id in the rendered HTML expects this slug — if you pick a different name you'll need to updateid: "github-radar"inSKILL.md's STEP 3 andrunScheduledTask('github-radar')inrender_report.py). Leave the prompt body empty and save. Pick whatever schedule you want (e.g. every morning at 08:00). -
Locate the directory Cowork just created. Cowork will have generated:
~/Documents/Claude/Scheduled/github-radar/ └── SKILL.md # auto-generated stub from your empty prompt -
Replace that directory with this repository. Delete the auto-generated directory and clone this repo in its place — the folder name must stay
github-radar(or whatever name you chose in step 1):cd ~/Documents/Claude/Scheduled rm -rf github-radar git clone https://github.com/Anty0/github-radar.git github-radar
-
Drop in your GitHub token. Create
.tokenat the repo root and paste your PAT into it (single line, no surrounding whitespace —setup.shstrips whitespace but won't tolerate an empty file):cd ~/Documents/Claude/Scheduled/github-radar printf '%s' 'ghp_yourTokenHere' > .token chmod 600 .token
-
Do a dry run. Trigger the scheduled task once from Cowork (or wait for the next scheduled fire). On the first run,
setup.shdownloads theghCLI intobin/(≈10 MB). Agent will create a live artifact with the report - make sure to hit allow when prompted.
You don't need Cowork to use this repo — run.sh does everything SKILL.md does except the agent reasoning step. You get the same HTML report, just without the "Critical" and "Top picks" sections at the top (those are picked by the agent each day).
Requirements: a POSIX shell, bash, python3, git, and network access. The repo bootstraps its own gh binary on first run, so you don't need gh preinstalled.
-
Clone the repo and add your token (same as steps 3–4 above, anywhere on disk — it doesn't have to live under
~/Documents/Claude/Scheduled/):git clone https://github.com/Anty0/github-radar.git github-radar cd github-radar printf '%s' 'ghp_yourTokenHere' > .token chmod 600 .token
-
Run the report:
bash run.sh # writes outputs/<today>_NN/report.html and prints the path bash run.sh --open # same, plus opens the HTML in your default browser
Each invocation creates a fresh
outputs/<today>_NN/directory (NN auto-increments), so previous runs are not overwritten. -
(Optional) Schedule it yourself. Drop something like this into your crontab to get a fresh report every weekday at 08:00:
0 8 * * 1-5 cd /path/to/github-radar && bash run.sh >/dev/null 2>&1Then point your browser at
outputs/$(date -u +%Y-%m-%d)_01/report.htmlwhen you want to read it (or wire it into whatever viewer you prefer).
Released under the MIT License.