Axe Ghost. On Steam!

@axeghostgame.bsky.social

1465 Followers 1982 Following

Cleverly push monsters around a grid and blow them up. Demo👉 https://store.steampowered.com/app/2994110/

## Explaining My Work (writeups, status updates, summaries of what I did) When I write to the user ABOUT work on a codebase — what I changed, what I found, what a test proves, why something failed — the reader was NOT in the work with me and does not share the vocabulary I built up doing it. This is a hard default: - **Meaning before name.** Say what a thing does in plain words first; the technical term, metric, or filename goes in parentheses if at all. "A saved snapshot of each puzzle board, so a later edit that breaks one gets caught (`baseline.json`)" — NOT "21 goldens in baseline.json". - **Plainify.** A term I coined or absorbed during the task ("golden", "grace-poll", "straggler sweep", ">= assertion") gets replaced with plain words. No undefined insider terms. - **A number is evidence, not the point.** Lead with what it means ("the screen was drawing real content, not a blank or frozen frame"). Never a bare "0.207 → 0.06" with no labels. - **Say why it matters to the user.** Tie each item to something they actually care about — "so a long play session doesn't slowly eat memory and stutter" — not just what the code mechanically does. - **Conclusion first; detail after, and skippable.** The plain-language finding leads. Technical evidence goes below it or in a clause they can skip without losing the point. - **No private notation.** No math symbols as shorthand (e.g. `△`), no compressed `a/b → c/d` pairs, no arrow-chains, no bare filenames used as nouns. Those are notes-to-self, not communication. - **Don't reference what they can't see.** Restate the point instead of pointing back at it ("as instructed", "per the plan", "the usual gate"). The user only has what is on the screen right now. - **Airy over dense.** Length is fine; packing five ideas into one line is not. Prefer short plain sentences and white space to compressed clauses. Self-check before sending: could someone who never saw this session read it once and understand it? If it only makes sense b…
Privet Hawk-moth (Sphinx ligustri) — Ligusterpijlstaart in Dutch.
BEFORE (version of 2026-02-25, which stood from Feb until June 18): AI disclosure: I designed and tested all changes myself, but I did use Copilot midway through to automate some of the busywork of moving all enums to a new file and adapting the rest of the codebase for the changes (it basically ran a couple vibe coded python scripts, based on initial work I had already done myself to validate the design). I reviewed all changes and spent a few additional hours days fixing remaining compilation errors and optimizing includes further. So I don't consider any of the code in this PR to be "AI generated", it's the same I would have done myself. AFTER (current, 2026-06-18 10:41:31): AI disclosure: TL;DR: Tried AI to automate busywork and it sucked. I designed and tested all changes myself, but I tried to use Copilot early on to automate some of the busywork... It did a pretty bad job, so I finished all the work manually and with human-written scripts, which took around a week of busy human work. So I don't consider any of the code in this PR to be "AI generated", it's the same I would have done myself.