0122 files · 164 KB
Agent desk
Give a standing job a plan, SQL memory and enforced limits, so your agent can pick it up tomorrow.
The library · 37 entries · 200 files
I’ve built the setup.Make it yours.
I give each standing job a desk, keep its memory in a database and teach my agents the work I want repeated. That gives them somewhere to pick up tomorrow, so I can spend my time on the decisions.
You can take the same desks, skills, stack and setup files, cleaned of anything private. Pick the pieces your work needs, so Claude Code or Codex gets one prompt to install your selection.
- 01
Pick your pieces
Add a desk, a skill or a whole shelf to your tray.
- 02
Leave your email once
Every prompt, file and zip opens on this browser. Copy one prompt for your selection.
- 03
Your agent installs it
Give the prompt to Claude Code or Codex. It shows you the plan before writing.
Let your agent help you choose
Read https://withseismic.com/library/llms.txt, list the entries on each shelf and ask which I want. Show your installation plan before writing, and ask before overwriting any file.01 · Desks
Give your agent a job to come back to.
I give each standing job a desk: its plan in desk.md, its memory in a database defined by SQL, its rules and its limits. The agent can work the daily round from that record, so it brings me the decisions instead of making me brief it again.
026 files · 35 KB
Business development desk
Keep companies, deals, calls and promises in one database, with the next decision ready for you.
033 files · 20 KB
Job desk
Keep roles, CV versions, applications and follow-ups together, with your approval before applying.
02 · The stack
Start with the account, not the blank folder.
I’ve put the account app, API and database into a starter, so your first brief can be about the product. Take the whole thing or borrow the parts that fit your repo.
015 files · 151 KB
The SaaS starter
Start with accounts, organisations, email and analytics wired into a React app and Hono API.
025 files · 6 KB
Turborepo
Run builds, tests and type checks across your apps in dependency order, with cached results.
0316 files · 18 KB
Hono
Give your API typed routes, generated docs and a consistent error response before adding features.
048 files · 8 KB
Drizzle
Keep your database schema typed and your SQL migrations checked in beside the code that needs them.
052 files · 5 KB
Better Auth with passkeys
Keep accounts, organisations and passkey sign-in in your own Hono API and Postgres database.
069 files · 10 KB
TanStack
Keep React routes, session checks and cached server data working from the same types.
075 files · 5 KB
Lefthook and Biome
Check staged code at commit time and types before a push, including work written by your agents.
085 files · 10 KB
Railway
Deploy the folder you intended, run migrations first and check the change on the live site.
095 files · 11 KB
Loops and Hogsend
Use Loops for account email and signups, or write bespoke lifecycle journeys in TypeScript with Hogsend.
106 files · 26 KB
PostHog
Define events once in TypeScript and build dashboards from those same names with a script.
03 · Skills
Teach the job once. Keep improving it.
I don’t write skills by hand: I tell Claude what I want, how it should look and what stack it’s on, so it writes the skill. I fix the instructions from real work, so the next run starts with what the last one taught me.
011 file · 6 KB
How I make a new skill
Tell your agent the job, the look and the stack, so it can write a skill you improve through use.
021 file · 9 KB
Tech lead
Keep a build moving from a plan in your repo to checks against the work running in production.
038 files · 68 KB
Carbon copy
Rebuild a live design from its DOM, styles and motion, with evidence for each section.
041 file · 13 KB
Design system
Turn repeated styles into shared tokens, typed components and written rules your agent can follow.
053 files · 24 KB
Hero writer
Give your landing page one clear visual idea, with a written brief and measured type and spacing.
062 files · 14 KB
Image parity
Build from a screenshot’s measured colours, type and geometry, with a tool to inspect the image.
0739 files · 419 KB
Lead magnet PDF
Turn sourced research into a branded PDF, with real interface captures and checks for overflowing pages.
082 files · 4 KB
Defer to Codex
Give Codex a full writing brief and have Claude check the result against the sources before applying it.
0920 files · 474 KB
Use Railway
Give your agent Railway’s own instructions for deploying services, reading logs and diagnosing databases.
104 files · 23 KB
Search data
Ask your agent for search performance, keyword demand and ad reports from the terminal.
111 file · 5 KB
Linear
Let your agent create and update Linear issues, projects and milestones through the command line.
04 · Guides
Give your agent the shape of the setup.
Voice, search data and mail all need accounts, permissions and somewhere to run. I’ve written down how I connect them, so your agent can work through the setup with your own details.
0114 files · 77 KB
Voice agent on Gemini
Let visitors talk through their needs, with call limits and the agent’s instructions fixed on your server.
021 file · 10 KB
Search Console and search volume
Connect Search Console and Keyword Planner to the terminal, so your agent can inspect your search data.
031 file · 5 KB
Claude Notch
See your Claude accounts’ limits beside the Mac’s notch and switch accounts without signing out.
044 files · 27 KB
Mail on Cloudflare
Give your agent a mailbox on your domain, with API access controlled through Cloudflare Access.
051 file · 5 KB
geo-kit
Keep schema, crawl rules and search measurement together, with shared helpers and agent skills.
05 · Software
Every app has a job.
I use Claude Code to run the work, Codex to write the words and Wispr Flow to get the brief out of my head. Each tool earns its place by changing how I work, so I’ve given you the reason to try it.
01Links
Claude Code
Put your agent inside the repo, where it can read the plan, run the skills and check its work.
02Links
Codex
Give the writing to the model I prefer for copy, while Claude checks every claim against its source.
03Links
Claude Desktop
Keep separate jobs open across Claude accounts, with Notch showing which account has room to work.
04Links
Wispr Flow
Speak the job, the look and the stack in one brief, so your agent gets the context together.
05Links
Cal.com
Carry a visitor’s voice-agent answers into a call booking, so they don’t have to start again.
06 · Builders
Turn the idea into a job an agent can finish.
I keep the plan, the acceptance criteria and the brief explicit, so the agent has something concrete to work against. Make your own in the browser; what you type stays there.
01Runs in your browser
Plan kit
Give your agent a plan in the repo: instructions, decisions, a backlog and the first PRD.
02Runs in your browser
Criteria checker
Find vague acceptance criteria and turn them into behaviours your agent can test.
See what I ship
I’ll send you one build log a week: what I added to the library and what I shipped with it. You can unsubscribe any time.