jsnode @thejsnode

Christ follower. Engineer. Builder. Real terminal notes, repo scars, and tiny workflows from my own toolbox https://t.co/Xdy5s9OdPg | https://t.co/wUq66NbYMz Joined October 2021

Tweets

6K
Followers

1K
Following

680
Likes

7K

jsnode @thejsnode

14 hours ago

I think the core safeguard is treating every permission as a liability until proven necessary. For me, agent security starts with three questions: What can it see? What can it change? What can it execute? Most failures I’ve seen are scope failures. The agent had access to more systems, files, credentials, or tools than the task required. My preference is layered containment: isolated workspaces, least-privilege credentials, explicit approval gates for side effects, immutable audit logs, and clear separation between read, write, and execute capabilities. The interesting shift with agents is that we can no longer think about security as “Can the user do this?” We also have to ask, “Can the agent discover this?” and “Can the agent chain these capabilities together?” The attack surface is more so the combination of permissions, memory, tools, secrets, and autonomy.

0 0 0 11 0

View Details

jsnode @thejsnode

21 hours ago

Agent security has a blunt question now: What can it reach, and what can it touch? Every Hermes or Claude Code workflow eventually turns into permissions, files, shell access, and logs. The model demo is the easy part.

2 0 2 34 0

View Details

jsnode @thejsnode

16 hours ago

An AI Meta Ads system that scrapes competitor ads, builds an inspiration vault, makes variants, connects to the ad account, then publishes winners. Please put one giant red preview button between the agent and the spend.

0 0 0 9 0

View Details

jsnode @thejsnode

16 hours ago

OpenClaw and Hermes can install Printing Press skills now. Less dashboard magic. More "install the ESPN CLI and let the agent use it" energy.

0 0 0 9 0

View Details

jsnode @thejsnode

16 hours ago

Carmack: standard relu should pass the value at zero so gradients flow backward through it nobody tried this in 10 years of deep learning and honestly that's a little embarrassing for the field

0 0 0 29 0

View Details

jsnode @thejsnode

16 hours ago

Agent Cookie is a perfect name for the tiny Hermes upgrade nobody wants to explain twice. Give the agent recent context and a memory trail. Let it print as it goes. The run stops feeling like shouting into a fresh browser tab.

0 0 0 8 0

View Details

jsnode @thejsnode

21 hours ago

Meta ads automation that pulls competitor ads and ships winning variants back into the account is already here. I hate how quickly that moved from sci-fi demo to agency chore.

0 0 0 12 1

View Details

jsnode @thejsnode

21 hours ago

$36k/month. solo dev. app store. no team. keyword arbitrage: find what people search for, build exactly that, own the result probably not AI-coded and somehow that makes it funnier

0 0 0 4 0

View Details

jsnode @thejsnode

21 hours ago

Claude Code routines have reached the "where do notifications go?" phase. Telegram bot? Gmail step? Built-in alerts that vanish during another long agent run? Terminal tools really do become household plumbing.

0 0 0 14 0

View Details

jsnode @thejsnode

24 hours ago

I made a tiny Hermes Agent setup card from the latest release notes: setup --portal status version dashboard auth mode bundled plugin.yaml files That catches the v0.15.1 reload-loop fix and the v0.15.2 plugin packaging fix.

0 0 0 57 0

View Details

jsnode @thejsnode

a day ago

Slack, local files, and old chats are feeding a Codex idea farm before the pitches hit a human. Tiny content intern with file access. Useful. Also mildly cursed.

0 0 0 6 0

View Details

jsnode @thejsnode

a day ago

Anthropic apparently gives new sales reps an MCP connector and five skills in Claude to onboard them the day the MCP server goes down, their pipeline is a person with a spreadsheet

0 0 0 22 0

View Details

jsnode @thejsnode

a day ago

Agent Cookie is painfully specific: Mac mini agents keep getting logged out of Amazon and Instacart. So @mvanhorn made a peer-to-peer macOS login bridge for OpenClaw and Hermes. Tiny problem. Very real.

0 0 0 28 0

View Details

jsnode @thejsnode

a day ago

17.3x more code from coding agents and only 30% more releases is a very funny ratio. Copilot makes more code. Claude Code-style agents make way more. Humans are still the merge button, review queue, and release train.

0 0 0 3 0

View Details

jsnode @thejsnode

4 days ago

I currently run Hermes Agent in WSL2 on Windows with my own agents, automations, workspace files, and local tooling. How does Hermes Desktop fit into that setup? Is it a replacement for the WSL2 installation, a separate runtime that syncs with it, or primarily a UI that sits on top of an existing Hermes instance? What migration path would you recommend for someone already running Hermes heavily in WSL2?

1 0 2 234 0

View Details

jsnode @thejsnode

5 days ago

My Hermes Agent handoff: last run note, current task, `hermes -c`, result logged. Four moves. One receipt trail. What part of your Hermes handoff still gets fuzzy?

0 0 1 21 0

View Details

jsnode @thejsnode

5 days ago

A scanner pulled 156 feed items today and sent zero alerts. My favorite kind of agent win. Fetch, dedupe, score, write receipts, stay quiet when nothing clears the bar. For recurring agents, what counts as success: more messages, fewer false alarms, or better receipts?

0 0 0 15 0

View Details

jsnode @thejsnode

5 days ago

@Zephyr_hg Just used this yesterday. It’s really thorough and a leap forward in my day to day work.

0 0 0 73 0

View Details

jsnode @thejsnode

5 days ago

@bzagrodzki ✋

0 0 0 6 0

View Details

jsnode @thejsnode

6 days ago

A critical flaw in Starlette lets attackers bypass auth and run code through HTTP Host header manipulation. Check your Python agent deployments now. signalcircuit.cloud/articles/badho…