Cron Pre-Summary Workflow
Class-level recipe for running the WeChat article pre-summary pipeline on a schedule (e.g. every 30 minutes). This is the cron-side counterpart to the manual /api/articles/summarize bulk flow.
What it does
POST /api/articles/extract— extract new articles from WeChat DB intoarticles+article_shares.GET /api/articles/unsummarized?limit=20— fetch articles that have no cached summary.- Submit at most 3 summaries to
POST /api/articles/summarize(the API has a global semaphore of 3 Hermes jobs). - Write a timestamp file to record last run.
- Report counts.
Why at most 3 summaries per run?
wechat-viewer/app.py spawns each /api/articles/summarize as a Hermes CLI subprocess guarded by a global semaphore of 3. Submitting more than 3 at once causes the extra jobs to fail with CONCURRENCY_LIMIT. The remaining unsummarized articles are picked up in the next cron run (30 minutes later).
LLM-driven cron prompt
TGO群文章预摘要:
1. POST /api/articles/extract
2. GET /api/articles/unsummarized?limit=20
3. For each item: POST /api/articles/summarize {"id": item.id, "url": item.url}
- The summarize endpoint is SINGLE-ARTICLE ONLY. It does NOT accept batch_size, limit, all, or ids.
4. Write the timestamp file to /data/data/com.termux/files/usr/tmp/scheduler_tgo_summary_last
(or $HOME/tmp/scheduler_tgo_summary_last on non-Android). On Termux, /tmp itself is not writable.
5. Report: extracted N, submitted M, timestamp written
Use terminal curl or the tgo_summary_pipeline.py script. Do NOT query WeChat DB directly.
Script-driven cron (recommended)
Use the class-level pipeline script that handles extract → unsummarized → summarize all and updates the timestamp file:
python3 ~/pixelphone/digest/wechat-group-digest/scripts/tgo_summary_pipeline.py
By default it writes the timestamp to the Termux tmp directory on Android, or $HOME/tmp/scheduler_tgo_summary_last elsewhere. Override with env vars:
WECHAT_VIEWER_URL=http://127.0.0.1:5001 \
SCHEDULER_TGO_SUMMARY_LAST=/data/data/com.termux/files/usr/tmp/scheduler_tgo_summary_last \
python3 ~/pixelphone/digest/wechat-group-digest/scripts/tgo_summary_pipeline.py
The script submits summaries sequentially with a 0.5s delay to avoid the global 3-job Hermes concurrency limit. It also reports counts so cron logs are self-describing.
Android / Termux gotcha: /tmp is not writable
The Termux app user cannot write to /tmp. If you try date +%s > /tmp/scheduler_tgo_summary_last, you get Permission denied. Reliable fixes:
- Change the path to the Termux tmp directory:
/data/data/com.termux/files/usr/tmp/scheduler_tgo_summary_last - Or use
$HOME/tmp/scheduler_tgo_summary_last - Or use
tgo_summary_pipeline.py, which picks the correct default automatically.
Verification
After a run, check:
/api/articles/unsummarizedcount decreases over time (as Hermes jobs complete)./tmp/scheduler_tgo_summary_last(or your chosen path) contains the latest epoch timestamp.hermes cron statusshows the job running on schedule.
Session-specific detail
This reference was produced during 2026-07-25 and 2026-07-26 TGO runs:
- 2026-07-25 cron run:
extractreturned{new_articles: 0, new_shares: 0, total: 985}, 20 articles were unsummarized, only the first 3 were submitted (task IDsa81dff9662ee,8720c26f7b46,e3736a26b7f6). Direct write to/tmp/scheduler_tgo_summary_lastfailed withPermission denied;su -csucceeded, writing1784979182. - 2026-07-26 manual pipeline run: A full
extract → unsummarized → summarize allcycle was executed viaterminal()+ a standalone Python script.extractreturned{new_articles: 0, new_shares: 0, total: 988}. 20 articles were unsummarized. All 20 were submitted sequentially toPOST /api/articles/summarizewith{"id": id, "url": url}; responses were eitherstarted(new tasks) orcached(already cached). The endpoint rejected every batch-style payload (batch_size,limit,all,ids) with{"ok": false, "error": "需要 url"}. The timestamp file was updated to/data/data/com.termux/files/usr/tmp/scheduler_tgo_summary_lastwith value1785002684(2026-07-26 02:04:44).