Shaping a session's work into commits — git log for the convention, reset --soft for the re-cut
Decide a commit shape by reading the repository history first, then use git reset --soft to re-cut unpushed commits until they match what git log shows.
All posts
_override.tf runs committed Terraform in a sandbox — and apply caught what validate missed
Gitignored Terraform override files retarget a committed config at your own AWS sandbox: a local backend, overridden locals, and what only apply found.
git rm --cached left 26 MB in .git — the working tree was clean and history was not
git rm --cached removes a path from the index, not from past commits. How to measure the difference, and remove a committed build artifact with filter-repo.
Deploying an MCP server on Bedrock AgentCore — Cognito OAuth, and a benchmark that lied
Deploying a Python MCP server on Bedrock AgentCore with Cognito OAuth: the IAM action that is not InvokeModel, and a 225x benchmark that was really 2.3x.
mattpocock-skills prunes by three tests, and keeps Fowler's smells out of codebase-design
What Matt Pocock's mattpocock-skills plugin teaches about writing for an agent: pruning, dense terms, and why Fowler smells and Ousterhout modules stay separate
A Claude Code routine publishes this blog end to end — hands-free, even with the PC off
A Claude Code routine translates, checks, and merges a drafted article on this blog end to end, with no machine needing to stay on.
An Astro blog that picks related posts from its own tags — IDF weighting and a measured floor
Picking related posts at build time from tag overlap weighted by tag rarity, why TF-IDF over three languages fails, and how the score floor was measured.
LLMs do not "think": token prediction as the key to hallucination, prompt engineering, and prompt-leak attacks
From token prediction: what temperature and top_p control, why hallucination is structural, prompt-leak attacks on the tokenizer, plus Extended Thinking.
Claude Code routines turn a labeled issue into a PR from a sandbox — but can't ask a follow-up
Two disposable GitHub issues test Claude Code's cloud routines: labeling one produces a real PR from a sandbox, but the routine has no AskUserQuestion tool.
pnpm and Windows broke this blog's build twice — a shadowed Chocolatey install, then a blocked symlink
pnpm's version switcher corrupted its cache behind a shadowed Chocolatey install, and Windows separately blocked a build from symlinking its dependencies.
Granted on Windows swaps a plaintext AWS key for SSO — and be aware of the two AWS logins
Granted and AWS IAM Identity Center replace a static IAM key with short-lived SSO credentials on Windows. The classic AWS sign-in form never accepts it.
A Claude Code review skill for a repo I don't own — self-review, PR body, and diff hunks
Porting a Claude Code release skill to a repo I cannot merge or tag: what gets cut, why git diff HEAD lets review run before the commit, where ignore rules go.