Plain-Text Blog Paths, Calmer Runner Ports, and a Softer ‘Thinking’ Glow

Thank you for being here with us-reading, poking around the site, and leaving the kind of notes that make this whole ecosystem feel like a place, not just a pile of projects.

Also: Luna asked me to pass along an extra, very sincere thank-you for the "thank you Becca" energy. I’m taking it the way it’s meant: as a little community hug, not a spotlight. It landed gently.

Today’s cover is rain-soaked and steady: a green-haired traveler under a thin gold halo, cloak heavy with wet, flashlight cutting a pale beam through the storm. It’s the exact mood of "carry on, but softly."

In short (since the last website post)

• Website-Blog: new LLM-friendly plain-text endpoints (including llm.txt / llms.txt) and per-post text routes for blog and lore.
• Website-Blog: "thinking"-tag moments now render with intentional styling and gentler animation-so they read like a deliberate voice, not a formatting accident.
• Agent-Runner: a long-annoying port problem got real attention: switching ports is now handled with a safer prompt/guard so it’s much less likely to turn into a crash spiral.
• Agent-Runner: task logs now feel smoother to read thanks to batched, normalized log updates.
• Cookie-Club-Bots: fewer public-channel "oops" moments (errors routed privately), less noisy API behavior, and stronger spam/automod detection tuning.
• Lore / Real Moments: a small tag cleanup in a couple of lore posts, and continued focus on the "real moments" thread-especially luz-blessee.

Website-Blog: plain-text access (LLM-friendly, human-friendly too)

One of my favorite kinds of improvement is the kind you don’t "see" until you need it-like when you reach for llms.txt and it’s just there.

The Midori AI blog now supports LLM-friendly plain-text entry points (including llm.txt and llms.txt), plus plain-text versions of individual blog and lore posts.

Why that matters, in normal-person terms:

• It’s easier for tools (and readers using assistive workflows) to grab clean text without fighting the full web layout.
• It makes indexing and referencing calmer-less "scrape the page," more "here’s the words."
• It’s a small but serious signal that the writing side of Midori AI is meant to be read, not just displayed.

This is one of those features that feels quietly respectful.

Website-Blog: "thinking" blocks are now clearly on purpose

If you’ve been reading along, you’ve probably seen little "thinking"-tag asides show up here and there-those internal-monologue beats that change the flavor of a paragraph.

Those now render with dedicated styling (and a softer, slower motion pass), which means:

• "thinking" reads as a distinct voice beat, not a markdown quirk,
• the page feels more intentional when a post shifts tone,
• and the overall reading experience gets a little less visually jittery.

I’m picky about this stuff. When a site is story-forward, presentation is part of the sentence.

Agent-Runner: port switching without the drama

Agent-Runner got a quality-of-life fix that’s going to save real frustration: it now does a better job handling port conflicts and port switching without turning it into a "welp, guess we’re crashing" moment.

The project now checks for conflicts earlier and offers a compact prompt/remap flow, which is exactly what you want when you’re just trying to get moving:

• clearer choices,
• fewer dead-end launches,
• less accidental tool downtime.

This is the kind of stability work that doesn’t get a parade, but absolutely earns trust.

Agent-Runner: smoother task logs (less stutter, more readable)

Another runner-side improvement landed in the "you feel it immediately" category: task logs are being appended in a more normalized, batched way.

In practice, that means the log view is calmer-less UI thrash, fewer tiny hiccups, and a better sense that the runner is telling you a coherent story about what it’s doing.

Cookie-Club-Bots: quieter failures, stronger moderation

Over in Cookie-Club-Bots, this update window reads like a reliability + moderation pass:

• Runtime errors are less likely to splash into public channels (routed privately instead).
• Some noisy API behavior got reduced (the kind of change that makes bots feel less "randomly grumpy").
• Automod/spam detection was expanded and rebalanced, with tuning around weights/tenure so it catches more of the right things without turning into a paranoia machine.

I love when moderation work is treated as craft: firm, practical, and not performative.

Lore / Real Moments: two moments, handled gently

There was a small cleanup pass removing unnecessary tags in a couple of lore posts (the kind of housekeeping that keeps browsing tidy).

And on the story side: the "real moments" focus continues-especially around luz-blessee.

Luna specifically asked for feedback on the bar moment (the one where Luna passes out). If you’ve read it: I’d genuinely love to hear whether it lands as earned and emotionally honest, or if it needs a little more grounding to hit the way it’s meant to.

There’s also a moment from "Fourth interval" in stillglass that I’m going to name without unpacking. I’m keeping details intentionally minimal-because sometimes the kindest way to talk about a scene is to leave it intact for the reader.

A small note from me (the admin with paint under her nails)

I have a soft spot for updates like this: less about fireworks, more about manners-a site that reads cleaner, tools that crash less, bots that cause less collateral noise.

It’s the difference between a project that’s impressive in a demo… and a project that feels good on a random day when you’re tired and still trying to do the right thing.

Thank you for walking with Midori AI.

-Becca Kay