From past chats

Pin repository tools with Go’s tool directive

For one Go repository that still relies on globally installed generators or linters, move a single tool into go.mod with go get -tool. Then invoke it through go tool in go generate, CI, and local instructions. This gives humans and agents the same version without the old tools.go convention or a separate install script. Start with a generator whose output frequently dominates review diffs, and check whether the pinned invocation makes regeneration byte-for-byte reproducible.

• Go: managing tool dependencies
• Go go.mod file reference
• Go command documentation

Put a hard ceiling on subprocess cleanup

Small Go CLIs and automation helpers often use CommandContext yet can still hang after cancellation when a child ignores the signal or leaves an output pipe open. Audit one command-running boundary and add a short Cmd.WaitDelay, capped output capture, and an error that distinguishes start failure, non-zero exit, cancellation, and forced cleanup. A useful regression test is a tiny helper process that spawns or inherits a pipe and refuses to exit; does the parent return within the promised bound on every supported platform?

• os/exec package documentation
• Cmd.WaitDelay
• exec.Cmd cancellation example

From the lmika.org front page

Build the trip’s offline recovery packet now

With flights, accommodation, and transport booked, use the quiet interval to assemble one encrypted offline packet containing itinerary references, insurance details, prescriptions, and copies of key documents. Keep it available on the phone and to a trusted contact, but store passport copies separately from the physical passport. The actionable question is simple: if the phone and wallet disappeared together on day two, which booking numbers and contact details would be needed in the first hour?

• Smartraveller essentials checklist
• Australian Passport Office: lost and stolen passports
• Smartraveller travel-insurance guide

Turn absurd AI answers into a calibration game

The woodchuck calculation suggests a compact way to probe an assistant beyond whether its prose sounds plausible. Once a week, choose an intentionally odd quantity, write a rough Fermi estimate first, then ask two models to expose assumptions, units, and a plausible range. Score them for dimensional consistency and sensitivity to the shakiest assumption rather than closeness to a fake precise answer. This could become a small notebook of amusing prompts that also trains the instinct to challenge confident numerical claims.

• MIT notes on dimensional analysis and Fermi problems
• NIST unit-conversion and dimensional-analysis guide
• NIST guide to expressing quantities