Free · weekly · Tuesdays

One real thing we built this week, with the file.

Not a roundup, not five links, not “10 prompts that will change your life”. One thing we actually shipped for a paying client, the file that did it, and what broke on the way.

/* ⚠️ hello@theclaudeguys.com — THE LONG DOMAIN, DELIBERATELY. Do not "fix" it to match claudeguys.com. brandcontext.md: "hello@theclaudeguys.com is a live mailbox on the long domain and does not move — never correct it." It was swept to the short form once, on 2026-08-25, by a domain pass reasoning from a frequency count (53 uses of the short form against 21 of the long) — which is exactly the kind of evidence that loses to an explicit written rule. D-1 governs the DOMAIN, not the mailbox. */Subscribe

No file, no send — if we didn’t build anything worth showing, you get nothing that week.

The rule we hold ourselves to

Every issue ships with a file

A real client problem

Anonymised where it has to be, never invented.

The actual file

The hook, the skill, the config — not a screenshot of one.

What it cost

Time saved, or tokens spent, as a number you can check.

THE ASSESSMENT

How good are you at Claude, actually?

Eight questions. You get the level you actually test at, the one move that gets you to the next one, how many hours a year the job you name is costing you, and three skill files cut to your title, company and industry.

Take the test

Under two minutes · no account · change any answer

  • 01The Beginner
  • 02The Chat Master
  • 03The Operator
  • 04The Automator
  • 05The Agent Conductor

Section surveyed 5,013 knowledge workers: 54% rate themselves AI proficient, about 10% test as proficient.

21 issues written

What’s coming

The schedule is public because we would rather be held to it. Each line below is written and queued, with the file it ships already in the library.

01

The four lines that stopped a DROP TABLE

A PreToolUse hook, the whole script, and why CLAUDE.md would never have caught it.

2026-08-11
02

Why your /goal never finishes

The evaluator never runs your tests. It only reads what Claude already showed it.

2026-08-18
03

Your 14 MCP servers are costing you 31%

We ran /context on a client's setup before they typed anything. Here's what it showed.

2026-08-25
04

Does the skill actually help? Prove it.

Same task, two runs, one number. The eval loop we run before any skill ships.

2026-09-01
05

400 of 460 lines were formatting

A PostToolUse hook that formats every file Claude touches, then hands lint errors back.

2026-09-08
06

31 hook events, six cadences, one map

Most courses teach eight. Here's how to pick the right one on the first try.

2026-09-15
07

Stop watching the terminal

Notification and Stop hooks that reach your desktop and your phone. Full script.

2026-09-22
08

We cut a $2,300 Claude bill to $890

Ten levers, ranked by what they actually returned. The first one is free.

2026-09-29
09

60 permission prompts an hour

The allowlist that ends the clicking without turning the safety off. Full JSON.

2026-10-06
10

Your Monday brief died when the lid closed

Routines vs desktop tasks vs /loop. One table, two questions, right answer.

2026-10-13
11

The $4 subagent that replaced a $70 one

Tool restriction plus a cheaper model. The whole agent file is 14 lines.

2026-10-20
12

What stops them turning the rules off

managed-settings.json is the only file a user cannot override. Full config.

2026-10-27
13

Claude wrote notes about you. Read them.

Auto memory is on by default, is not CLAUDE.md, and is usually wrong somewhere.

2026-11-03
14

61 skills installed. None of them fired.

The 1,536-character budget, the 1% listing cap, and a linter that finds the dead ones.

2026-11-10
15

The hook that worked on one machine

Exit 2 or JSON, never both. Plus a probe that shows you every event's payload.

2026-11-17
16

Move the PR reviewer off the laptop

claude -p in CI, with a read-only tool list, a turn cap and a cost line. Full YAML.

2026-11-24
17

Can it read the credentials file?

Three layers of secret protection, and the one everybody leaves open.

2026-12-01
18

Hour three, Claude forgot hour one

A PreCompact hook that writes the handoff and a SessionStart hook that reads it back.

2026-12-08
19

Your CLAUDE.md is 640 lines. Cut it.

Memory, skill, rule, or hook? One question routes every instruction correctly.

2026-12-15
20

The CISO wanted evidence, not assurances

Five artifacts that turn a Claude rollout from a conversation into a review pass.

2026-12-22
21

Claude writes the script that runs the agents

Dynamic workflows, the 1,000-agent cap, and the resume rule that costs you.

2026-08-18

Issues become readable here on the morning they send. The files are already in the library if you would rather not wait.

NO PITCH

Start with the free thing. Decide about us later.

The library is ungated and the newsletter is free. If neither is useful, nothing else we sell will be either.

Get the starter pack →