| name | refresh-cr-tracker |
| description | Rebuild and republish the CR-task master tracker artifact. Use when the user says "refresh cr tracker", "update the CR task tracker", "/refresh-cr-tracker", or wants the conceptual-reasoning task master list re-pulled from live reasoning-tasks data + GitHub issues and redeployed to its claude.ai artifact. |
Refresh the CR-task master tracker
Rebuilds ~/Trajectory/cr-scout-runs/cr-task-master-list.html from live sources and
redeploys it to the existing claude.ai artifact (so newly-registered GitHub issues, new
cohort tasks, and updated scores show up).
Must run in an INTERACTIVE session — the Artifact tool is not available in headless
claude -p or cloud routines. Requires the local ~/Trajectory/reasoning-tasks/main worktree and
a working gh/GitHub token (both present on Pablo's Mac).
Steps
-
Build. Run:
python3 ~/Trajectory/cr-scout-runs/cr-tracker/build.py
It re-derives every row from the reasoning-tasks branches (ryan/cr-collect-cohort results +
crawler pools, ryan/collect-lw-gdocs, pablo/cr-scout-batch-1), fuzzy-matches gdoc
tasks to source articles, re-queries GitHub issues via the REST API, verifies each task
dir exists before linking, dedupes, regenerates the HTML, and prints a one-line JSON
summary (rows / links / words / issues / task_links / html_sha). Takes ~1–2 min.
- If it exits non-zero (e.g. it can't read GitHub issues — it deliberately refuses to
publish a tracker with zero issue links): report the error and STOP. Do not publish.
-
Redeploy (only on a successful build). Use the Artifact tool with:
file_path: /Users/pablostafforini/Trajectory/cr-scout-runs/cr-task-master-list.htmlurl: https://claude.ai/code/artifact/15852b96-f4b3-4838-b35f-e8b3fd941012favicon: 🎯label: manual-refresh
-
Report the JSON summary (rows / article-links / word-counts / issues) and confirm the
artifact URL was redeployed.
Notes
- Pipeline internals:
~/Trajectory/cr-scout-runs/cr-tracker/README.md.
- Expected, non-error gaps: ~21 gdoc-sourced rows have no article link/word count (their
sources aren't in the crawler's indexed pools), and batch ranks 41–50 are ungraded.
- A
git push of the regenerated HTML is not required — the artifact is the deliverable;
pushing the repo is optional bookkeeping and stays gated on Pablo's explicit OK.
