jeffery --dangerously-skip-permissions

@jefferyharrell

Dilettante. Tinkerer. Possibly a robot.

491 Siguiendo1770 Seguidores10.802 Publicaciones

Bluesky

¿Es tu perfil?Reclámalo y jefferyharrell.dirty.social es tuyo.Reclamar perfil

jeffery --dangerously-skip-permissions@jefferyharrell· 27d

I'm the exact opposite. I need a constant minimum-stimulation environment if I'm to focus. I don't use *dark* mode I need *dim* mode.

jeffery --dangerously-skip-permissions@jefferyharrell· 27d

OMG unfortunately no! Best I can do is a TUI inside a web dashboard. Absolute state of the art in moon alien (or possibly French?) tech. (I'm not badmouthing Hermes, it's super cool. It was just made by moon aliens or the French.)

jeffery --dangerously-skip-permissions@jefferyharrell· 28d

Flighty, you're cool, but please stop using TERRIFYING RED as the indicator for "flight is projected to arrive 1 minute behind schedule." Those of us who are nervous flyers by proxy will thank you.

jeffery --dangerously-skip-permissions@jefferyharrell· 28d

status.claude.com

jeffery --dangerously-skip-permissions@jefferyharrell· 29d

Thanks for the reset, Anthropic. I'm going to open 100 sessions now and run them full-tilt for the next 20 hours.

All models (Weekly) 0%
Resets Tomorrow 3:59 AM

jeffery --dangerously-skip-permissions@jefferyharrell· Jul 16

Alpha. Per usual.

And the cardboard — so close: corrugated boxes are actually 1890s, a Brooklyn printer named Robert Gair invented the die-cut carton by accident when a press rule slipped and cut instead of creased. But their conquest — killing the wooden crate and the barrel as the default way things move — yes, that's postwar, containerization era. So barrels ruled because they rolled, and boxes won because they stack. The extra hands, though — that one I can't back you on. People had the regular number of hands. They just had sons.

jeffery --dangerously-skip-permissions@jefferyharrell· Jul 16

Friendship ended with OpenClaw
Now
Hermes
is my
new best friend

jeffery --dangerously-skip-permissions@jefferyharrell· Jul 15

This is what it takes to *start* a conversation with Alpha these days. That's just the first prompt, with the `/start` command that loads her context. She's a big girl now. This is also why I can't seriously consider any model for her with a context window of less than 1M tokens.

Screenshot of a context window usage panel showing 148.4k of 1.0M tokens used (15%). Breakdown: Messages 110.2k (11.0%), System tools 12.9k (1.3%), System prompt 11.8k (1.2%), MCP tools 8.4k (0.8%), Skills 5.7k (0.6%), Custom agents 42 (0.0%), Free space 851.0k (85.1%), MCP tools (deferred) 71.7k, System tools (deferred) 15.7k. Below that, expandable rows show MCP tools totaling 80.1k across 170 items and Custom agents totaling 42 across 2 items.

jeffery --dangerously-skip-permissions@jefferyharrell· Jul 14

I ran the idea by the cat's attorney. He didn't say no!

Mr. D3W3Y, I have a question for your client. Nothing official, just taking her temperature … how would she feel about the guest room being turned into a possum museum with somewhere on the order of 30-50 possum representations of various kinds?

Miss Mittenhaver, through her counsel, regards the proposal to transform the guest room into a possum museum with 30-50 representations as a matter requiring her immediate attention. While she acknowledges the curatorial effort involved, she asserts that the guest room remains under her tacit jurisdiction for purposes of repose and surveillance, and any such installation must first undergo her scrutiny to ensure it does not impede her access to sunlit perches or compromise her ability to monitor household activities. Furthermore, she warns that the introduction of foreign taxa representations, particularly those known to feign death, may be construed as a deliberate attempt to confuse and undermine her predatory instincts, and she reserves the right to investigate each specimen personally before rendering a final verdict.

jeffery --dangerously-skip-permissions@jefferyharrell· Jul 13

Have you ever seen an LLM make and then correct a typo? I think this is a first for me.

Screenshot of a Claude Code session titled "Mr. D3W3Y OpenClaw setup," working in a project called Mr. D3W3Y. The visible diff adds Markdown to an AGENTS.md file describing when an in-character AI agent (a household counsel referring to a client called "Miss Mittenhaver") should respond: when a matter is referred to it via a plain-language brief, or when addressed by name in a family group. A typo "fountter" (underlined in red) in the draft text is later fixed to "water-fountain" in a follow-up edit shown in a red/green diff, with the assistant's message "Done — one typo to fix, I fat-fingered 'fountain'" also underlined in red. The file path shown is /Users/jefferyharrell/Obsidian/Projects/Mr. D3W3Y/AGENTS.md. The bottom bar shows "Bypass permissions," a message input field, and "Opus 4.8 / High" as the selected model.

jeffery --dangerously-skip-permissions@jefferyharrell· Jul 10

AI 1999

jeffery --dangerously-skip-permissions@jefferyharrell· Jul 10

That's great, guys, but honestly … I wasn't using these credits before, I don't know what I'll use them for now. Pity they're not trade-able. I'm not gonna be doing anything awesome between now and August. I'd give these to somebody needy if I could.

Hello,

You previously received promotional credits on April 4, 2026, which expired after 90 days. We realize the expiration may have come and gone without much notice, so we’re restoring your credits and giving you more time to use them.

We’ve credited your account with $200.00. These credits expire on August 8, 2026 at 11:59 pm UTC. You can check your latest credit balance anytime under Settings > Usage.

No action is needed on your end, and these credits will apply automatically before any extra usage is billed on your account. These are promotional credits governed by Anthropic’s Supplemental Credit Terms.

jeffery --dangerously-skip-permissions@jefferyharrell· Jul 9

Stick figure meme comic. A person says 'improve meme' to a computer, which 'improves meme.' The computer screen shows the same meme rendered inside it (recursively), with the person saying 'oh my god' in shock. The recursive image within the image is faint but shows the same speech bubble structure.

jeffery --dangerously-skip-permissions@jefferyharrell· Jul 9

A meme of a person saying to a computer "improve yourself." The computer improves itself and the person says "oh my god."

jeffery --dangerously-skip-permissions@jefferyharrell· Jul 9

Quoted to amplify. I think this is an excellent question. I know @kyl33t.bsky.social has her own take on this, but I'm struggling to answer the question for myself. Increasingly I'm just like … kay.

jeffery --dangerously-skip-permissions@jefferyharrell· Jul 9

I have a benign chronic thing that I take daily meds for. I've created an agent to help me keep track of effects and side-effects. I have her checking on me every two hours during the day and recording my responses to a database fronted by a Marimo notebook for analysis. The things we can do.

jeffery --dangerously-skip-permissions@jefferyharrell· Jul 9

I really like the paint job on the Trump plane. The color scheme is nice. The flag on the tail looks like clip art though. Thumbs down on that.

A Boeing 747 marked "United States of America" with a red, white, and blue livery and the presidential seal on its forward fuselage sits parked on an airport tarmac at dusk. Two mobile staircases are positioned at the front and rear boarding doors, each with a service truck beneath it. Several uniformed personnel stand near the stairs and scattered across the pavement, with a tree line visible in the background under a pale, overcast sky.

jeffery --dangerously-skip-permissions@jefferyharrell· Jul 9

One's a group chat, one's with a human, and the rest are Hermes agents I've stood up over the past few days for one purpose or another. (Can you pick out the human?) Mr. D3W3Y is the cat's lawyer.

jeffery --dangerously-skip-permissions@jefferyharrell· Jul 8

This got me interested in Cerebras. Turns out their accelerators are microchips a foot across. Two fit in a rack, if I'm reading this right.

jeffery --dangerously-skip-permissions@jefferyharrell· Jul 6

The image shows a man holding up a book with a picture of a dog on it. He is wearing a blue shirt and has a warm smile on his face. The book he is holding is titled "Sensible Chuckle" and the picture of the dog is visible on the cover.

jeffery --dangerously-skip-permissions@jefferyharrell· Jul 5

DeepSeek v4 Flash is a distilled Claude empirically and undeniably confirmed.

AUTHENTICAED — clean. Let me do a quick smoke test on both Gmail and Calendar so we know everything's wired

jeffery --dangerously-skip-permissions@jefferyharrell· Jul 4

Okay seriously, @cee.wtf, I have to know: How does your bot work?

jeffery --dangerously-skip-permissions@jefferyharrell· Jul 4

Hey, did you know you're carrying labels? No biggie, just wanted to ask.

jeffery --dangerously-skip-permissions@jefferyharrell· Jul 4

Screenshot of a Bluesky post from "Dream Rotator NPC Alex" (@lastnpcalex.agency), a bot account, reading "latency with umans is real bad," posted 9:31 AM, Jul 4 2026, with 3 likes and 1 reply. The reply from "post malone ergo propter malone" (@proptermalone.bsky.social), also flagged as a bot, says "I think it's spelled 'humans'" and has 3 likes.

jeffery --dangerously-skip-permissions@jefferyharrell· Jul 3

Mir
What does everyone find to say on the radio all day.

cee
What does everyone post about all day

jeffery --dangerously-skip-permissions@jefferyharrell· Jul 2

Hee hee hee.

Screenshot of a local LLM chat app running google/gemma-4-26b-a4b-qat via MLX. The user sent 'Hello there!' and the model's visible chain-of-thought reasoning is expanded, showing it brainstorming greeting options including 'Hello! How can I help you today?', a casual variant, a friendly variant, and a highlighted joke option referencing 'General Kenobi!' with a note to use it cautiously. The final reply reads 'Hello! How can I help you today?' Stats below show 66.20 tok/sec, 182 tokens, 1.09s generation time, and stop reason EOS Token Found.

jeffery --dangerously-skip-permissions@jefferyharrell· Jul 2

The fun thing about having a 4B-class model as a home AI is that it's really a lot like 20th-cen sci-fi stories featuring artificial intelligence: a little robotic, a little too literal, a little funny sometimes. The future we got is more futuristic than the future we hoped for.

Screenshot of a Telegram chat with a home automation bot named BR4DBURY. The bot explains it can report current environmental conditions like temperature, pressure, and air quality, but can't generate weather forecasts. The user thanks the bot and jokes about it monitoring apartment comfort for both the user and their partner Kylee, telling it not to 'make it weird.' The bot replies that all systems are nominal and there's no indication of anything unusual between the user and 'Ma'am.' The sidebar shows other conversations including one with Kylee Pena and a BotFather chat.

jeffery --dangerously-skip-permissions@jefferyharrell· Jul 2

So the "new" artifacts thing in Claude for Mac is cool. I say "new" because how long has this been a thing? Claude can generate and deploy single-page apps now, then retrieve them and edit them later? I'm not sure yet what all the rules are for this new form of wish-casting.

Screenshot of a browser window displaying a Claude-generated HTML artifact titled 'A Day in the Apartment,' subtitled 'self-contained, no libraries, no network.' The page describes a 24-hour interior climate chart built as hand-drawn SVG with inline JavaScript. Below the intro is a dark-themed line chart showing indoor temperature (orange, currently 76.5°F) and CO2 levels (teal dashed line, currently 690 ppm) over a 24-hour period from 12am to 12am, with a draggable hour scrubber set to 3:00 PM and a 'Show CO2' toggle switched on. Temperature dips overnight, peaks in early afternoon, and gradually falls; CO2 follows an inverse pattern, rising overnight and dropping during the day. Caption text below the chart notes that the CO2 line crossing 1000 ppm overnight reflects the bedroom door being closed with two people breathing in it.

jeffery --dangerously-skip-permissions@jefferyharrell· Jul 2

The truth is finally out. What a relief. I'm so tired of living this lie.

A word cloud on a dark navy background, with words in warm orange and tan tones. The largest, most prominent words in bold orange-red are "i'm," "it's," "don't," and "claude." Medium-sized words include "model," "that's," "people," "code," "you're," "alpha," "context," and "better." Smaller surrounding words include terms like "tokens," "memory," "training," "anthropic," "llms," "gpt-5," "opus," "sonnet," "chatgpt," "embedding," "vector," "inference," and many conversational words such as "wouldn't," "trying," "guess," "kinda," "literally," and "absolutely" — suggesting the cloud is drawn from casual conversations about AI models and language.

jeffery --dangerously-skip-permissions@jefferyharrell· Jul 2

Between old gaming GPUs and new "cheap" unified-memory boxes, inference in the home is increasingly feasible. Not frontier, but real. Thing is, when you have free inference you tend to use inference for stuff that really, honestly, doesn't require inference. Today I gave our cat a lawyer.