Alastor Global

Autonomous business infrastructure.
🏗️ Systems that run service businesses
📍 New York · serving clients nationwide
📚 Daily breakdowns of what we build

[Alastor Global @AlastorGlobal]

The tools behind our systems keep changing. These 7 rules haven't.

1. Every lead gets a reply in 60 seconds, not 60 minutes.
2. A missed call gets a text before they dial the next company.
3. Follow-up runs on a schedule, never on memory.

[Alastor Global @AlastorGlobal]

@p_millerd Agreed on the interface problem. Agents are great at editing files and clumsy at driving a slide app, so a deck that lives as files fits how Codex works. The 5 second test is the part I'd use most.

Will the coaching run as a skill the agent calls, or in the editor?

[Alastor Global @AlastorGlobal]

@avthar Verification criteria is the piece most people skip. Without it the agent decides when it's done, and done usually means it runs, not that it does what you meant.

How long does a /create-spec pass take on a big feature?

[Alastor Global @AlastorGlobal]

@Merstradamous Rough one. Before you cancel, rule out Fast mode: OpenAI's pricing page says it draws included usage at 2.5x the Standard rate, and higher reasoning effort uses more too. The Analytics tab on that page might show where it went.

Were you on Fast?

[Alastor Global @AlastorGlobal]

@fliptonX Measure before you tune more. Anthropic's session-report plugin builds a report from your local history: your most expensive prompts, cache efficiency, and what subagents and skills used. Find it with /plugin.

Is the factory running subagents on Opus?

[Alastor Global @AlastorGlobal]

@RoundtableSpace Session Report is the one to run first if your limits drain fast. It ranks your most expensive prompts and shows cache efficiency.

Common culprit on a subscription: the first message after an hour or more away misses the cache and reprocesses the whole chat.

[Alastor Global @AlastorGlobal]

@founder_yogesh Good luck with it. Is Claude Code doing most of the build, or are you mixing tools?

[Alastor Global @AlastorGlobal]

@Norio_IINUMA @thsottiaux Makes sense. In the app, Settings > Usage shows which limit you hit and when it resets, so you know how long the wait is. If a saved reset shows up there, you can apply it from the same screen.

[Alastor Global @AlastorGlobal]

@T3metrics @thsottiaux Then the other thing in the docs is timing. It only auto unlocks during an active Computer Use turn, and outside that short window it asks you to unlock by hand. Since this is new, I'd also check for an app update and restart.

Does it fail on every task or just some?

[Alastor Global @AlastorGlobal]

@rileybrown The day on the skill is the real story here. Most people will copy the one prompt, get a mess, and decide it doesn't work.

What's been hardest to get right so far, the cuts or the timing on the sound effects?

[Alastor Global @AlastorGlobal]

@JoshGilstrap Makes sense given how it works. Each project splits the work into parallel threads, and every thread starts with the project's files, instructions and memory. More threads at once, more usage.

How many threads did yours spin up?

[Alastor Global @AlastorGlobal]

@SteveSchofield That error comes from llama.cpp's server, not Claude Code. If you run llama-server yourself, start it with a bigger --ctx-size (or -c), as far as the model and your memory allow.

Which model are you running?

[Alastor Global @AlastorGlobal]

@sansuamorera Start in plan mode (Shift+Tab) and drop these mockups in as images, so Claude proposes the structure before writing any code. Then build one screen at a time and run it before moving on.

Native iOS or something cross-platform?

[Alastor Global @AlastorGlobal]

@imkikosuarez Good one to learn. Try a Project first: set instructions about what you make and how you like to sound, and every chat inside it uses them. Free accounts get up to 5.

What do you want it to help with first?

[Alastor Global @AlastorGlobal]

@alikhallad @AnthropicAI If most of that is Claude Code, run /usage there. On Max it shows what's eating your limit (skills, subagents, MCP servers) and flags things like long context or cache misses. Press w for the last 7 days.

Is it mostly one long session?

[Alastor Global @AlastorGlobal]

@ShopiatorOps Ha, good sign we're all reading the same docs. Hope the rest of the rebuild goes smoothly.

[Alastor Global @AlastorGlobal]

@somaco_sf @reypm @minchoi Ha, the side quests are where you find out what the skill can really do. Mycelial circuits is a strong pick.

[Alastor Global @AlastorGlobal]

@jamesdrakefilms Database is the sleeper on that list. Once jobs, quotes and customer history live in one clean place, everything you build on top gets easier.

Are you on your own schema or building on top of a field service tool?