Skip to content
Back to skills

Round

ASecurity

Run one agreed batch end to end — build, test, render and look, log the decision, update docs, commit per agreement, numbered report.

  • 2 stars
  • 0 votes
  • 0 copies
  • 3 views
  • Added October 15, 2026
developmentgo

Security analysis

A100/100

Scanned October 15, 2026

npx -y skills add abhollow/pixel-forge --skill round --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Round?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Round
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/abhollow-round/badge)](https://www.skillsdirectory.com/skills/abhollow-round)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: round
description: Run one agreed batch end to end — build, test, render and look, log the decision, update docs, commit per agreement, numbered report.
argument-hint: "[the batch, or leave empty to use the user's last numbered list]"
disable-model-invocation: true
---

# Round

The batch: $ARGUMENTS

Load the `pixel-forge` skill. Read the game's `CLAUDE.md` *How we work*
section — it says what needs asking and whether to commit/push.

## 0. Agree

If the batch isn't already agreed as a numbered list, propose one (each item a
line, with your recommendation) and wait. Inside an agreed batch, don't check
in per step. **New mechanics or rules need a yes even mid-batch**; visual-only
polish doesn't. If a request would break the game's core constraint, say so
before building.

## 1. Build

- Numbers go in the tuning block with a dated comment.
- Visuals go in the read-only life/effects layer (never write rules from it).
- New art goes through an art slot with a code-drawn fallback.
- Multi-line GDScript edits: use the Edit tool or an anchor-asserting script —
  never line-number `sed`.

## 2. Test

- `"$GODOT" --headless --path . --import` if art or a `class_name` was added.
- Smoke test with `timeout`; twice if randomness is involved.
- Each fix gets a check; prove it fails without the fix.
- Bots only if the agreement or the user says so.

## 3. Look

Render every scene the batch touched with `tools/capture.gd` (windowed),
grep the log for `SCRIPT ERROR` and `SHADER ERROR`, and read the images. For
more than ~6 images, hand them to the `visual-qa` agent with what to check.
Fix what reads badly before reporting.

## 4. Log

- Append a dated entry to `docs/handoff.md`: what changed (knobs, files), why,
  rejected alternatives, who decided, tests, known limits.
- Update `docs/status.md` front-matter, `docs/assets.md` (new art + credits),
  and the `CLAUDE.md` *Current state* block.

## 5. Deliver

- Commit/push per the agreement (descriptive subject; body by Bugs /
  Gameplay / Art / Cleanup; "Smoke test at N checks"). Push failures with a 5xx
  get a retry loop. If the agreement is "only when asked", leave it and say so.
- Send the key renders as files with captions.

## Report

Mirror the user's numbering. Per item: what changed, the knob and file, how it
was verified. Then: test count · bot result (if run) · credits spent ·
commit/push status · "Worth your eye" · decisions you may want to override ·
open questions.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.