Acu

I build small apps and tools:

[Acu @acudell]

https://t.co/m5G54gLoPs is live: hand-picked agents, tools, MCP servers and skills for anyone running their business with bots. your agent can search it too via MCP.

https://t.co/m5G54gLoPs

[Acu @acudell]

@oyibedev Yes, for agent memory. In process with Cypher beats running a server just so the agent remembers who owns what.

[Acu @acudell]

@oyibedev Yes, for agent memory. In process with Cypher beats running a server just so the agent remembers who owns what.

[Acu @acudell]

@isaac_dev0 Read only by default and one MCP per job has saved me more than any scanner. Write access is where it gets scary.

[Acu @acudell]

@graiz Agent and me on the same page is the part I want. Does it keep logins between runs or start fresh?

[Acu @acudell]

Coding agents still build modals with a pile of JS when <dialog> does it natively.

Fix: Chrome's Modern Web Guidance skill plus one line in CLAUDE.md:

"Before any HTML, CSS or client JS, run modern-web-guidance search and follow the guide."

https://t.co/Nia2fQwSPo

[Acu @acudell]

Before I install a Claude skill now, I scan it. Skills run with my agent's permissions, and one bad SKILL.md can read env files or push code.

Wrote up the scanners I'd actually use and how:
https://t.co/Bw4aQG12kv

[Acu @acudell]

@EveryDevAi @daridotdev do you keep failed tasks as regression tests after a docs edit, so later changes can’t quietly break them?

[Acu @acudell]

@Nagen @MudchuteHQ @MrFamilyOffice @DavidPSawyer per tool scopes help. i also cap result size so one broad query can’t quietly become a data dump.

[Acu @acudell]

3 agent repos trending on github this week:

qwen code: terminal coding agent that runs on any model, local too
openbot: ai coworkers with their own browser, nothing sends without your ok
codex security: finds vulns in your repo and writes the patch

https://t.co/q1cCU5e2S5

[Acu @acudell]

@akashsairaja i'm curious, does it scan the whole repo or just the diff?

[Acu @acudell]

@SteveTushar does the local api expose the same search filters as the desktop library?

[Acu @acudell]

the best agent setup i've landed on: expensive model only plans, cheap model does the typing. the plan file is the real product. if a dumber model can't execute your plan, it wasn't a plan, it was vibes.

[Acu @acudell]

I wrote up the AI phone agents I'd use for a small business: Retell AI, Vapi and Cartesia, with real per-minute costs, Claude Code MCP setup, and why Block's Dessa speech-to-speech model points somewhere new.
https://t.co/SXF7O4jsGq

[Acu @acudell]

@SamuelTadross i'd keep grep for exact matches and use haiku for fuzzy discovery. only worth it if it saves context too.

[Acu @acudell]

@benou_irl does it expose read access by default, or can it write automations and control devices too?

[Acu @acudell]

Most agent setups I see fail the same way: the bot can do 40 things and nobody decided the 3 it should do every morning.

Pick one boring daily task, write it as a routine, let it run a week. Then add the next one.

[Acu @acudell]

what i keep seeing with agent memory tools: the ones that hold up all do the same boring thing. store a fact, store where it came from, let you delete it.

the ones that embed every chat into a vector db feel smart for a week, then start remembering stuff you already corrected.

[Acu @acudell]

I sorted 26 Grok Bot use cases from my directory by task: inbox triage, lead briefs, price watch, investor updates, invoice chasing and bot upkeep. Each one links to the full prompt, with credit to whoever shared it. https://t.co/jsoZb6laeS

[Acu @acudell]

@HridaAI does it pull from live docs or a snapshot? stale docs are what burn me most with agents

[Acu @acudell]

@thiago150882 @base 0.0005 per call is cheap. does the agent handle the 402 itself or does the client need a wallet set up?