What you get
Three times a day, a queue of drafts lands in your messages. Below is one morning’s delivery.
Each run either says nothing cleared the bar, with its search and extract counts, or delivers drafts. Every draft carries its sources, a claim-by-claim fact-check and a style score. The samples below show the range. One is fully sourced and scores 94. One has a claim (“shipping next month”) that appears in a single maintainer’s post, so the draft hedges it and the fact-checker suggests cutting it. One scores 79, under the 85 bar, for two hedges in the lede and the banned word “landscape”.
(The stories are invented. Tap a collapsed draft to expand it.)
Hermes v0.19.0 lands: what “Quicksilver” changes
Dek: Nous Research tagged v0.19.0 on 20 July. Here’s what matters if you run it as a daemon.
[draft body: 420 words, structured per /house-style: lede, two sections, further-reading]
- ① github.com/NousResearch/hermes-agent/releases, tag v2026.7.20
- ② hermes-agent.nousresearch.com/docs, changelog page
- ③ to ⑤ [two maintainer threads via x_search, one docs diff]
| “tagged 2026-07-20” | ✓ source ① |
| “219k GitHub stars” | ✓ source ① (checked live) |
| + 4 more claims | ✓ all matched |
Two agent frameworks are quietly merging their skill formats
Fact-checker note: claim 3 (“shipping next month”) appears only in one maintainer’s post, with no second source. The draft hedges it. Consider cutting.
Weekend read: what “metered tool gateways” cost
Style note (self-scored): two hedges in the lede, one banned word (“landscape”). Below the 85 bar, so it needs an edit pass before approval.
Approve, Edit and Spike are replies in the chat thread. Nothing here posts anywhere.
Why this shape
- The draft cites a numbered list of claims and URLs, so you never have to ask where something came from.
- The Fact-Checker never sees the draft. It gets the claims and URLs and fetches each source itself. (Trap 2 says why.)
- The score comes from the rubric in your own house-style
SKILL.md. When it’s wrong, you fix the rubric and the fix sticks. - Quiet runs report their work. A pipeline that pads quiet days is worse than none.