Armory — Standalone Claude Code Skills

Standalone Claude Code skills forged from real problems.

Every skill in this repo started as a problem I watched people work around instead of solve. I built the structural fix, tested it, and published it here. No frameworks, no dependencies, no setup beyond a folder copy. Each skill works on any codebase, on any OS, the moment you drop it in.

Install

Prerequisites: Claude Code

Clone the repo, then copy any skill folder into your project:

git clone https://github.com/SethGammon/Armory.git

Mac / Linux:

cp -r Armory/systematic-debugging your-project/.claude/skills/

Windows (PowerShell):

Copy-Item -Recurse Armory\systematic-debugging your-project\.claude\skills\

Windows (Command Prompt):

xcopy /E /I Armory\systematic-debugging your-project\.claude\skills\systematic-debugging

Or drag the folder in Explorer. There is no installer — the skill is the folder.

Skills

Skills marked with † are also included in Citadel. If you have Citadel, you already have them.

Debugging & Review

Skill	What it solves
systematic-debugging †	Prevents shotgun debugging. Forces 4-phase root cause analysis — observe, hypothesize, verify, fix — with an emergency stop after 2 failed fix attempts.
review †	5-pass structured code review: correctness, security, performance, readability, consistency. Language-agnostic.
triage †	GitHub issue and PR investigator. Searches the codebase for root cause, proposes fixes with file:line references, optionally implements them.

Code Generation

Skill	What it solves
scaffold †	Generates new files by reading 2+ existing exemplars first. Output matches your project's conventions because it is derived from them.
refactor †	Safe multi-file refactoring with automatic rollback. Establishes a type/test baseline before touching anything, verifies zero regressions after.
test-gen †	Generates tests by reading your project's own framework and patterns. Happy path, edge cases, error paths.
doc-gen †	Generates documentation by reading existing style first and matching it. Three modes: function-level, module-level, API reference.

Research & Optimization

Skill	What it solves
research †	Converts a question into structured findings with confidence levels and source citations. Produces information, not decisions.
research-fleet †	Parallel multi-scout research. Decomposes a question into independent angles, runs scouts simultaneously, compresses findings between waves.
experiment †	Automated optimization loop. Proposes changes in isolated worktrees, measures with a metric command, keeps improvements, discards failures.

Diagrams & Utilities

Skill	What it solves
ascii-diagram	LLMs misalign ASCII diagrams because they predict tokens sequentially. This skill uses a programmatic character-grid — alignment is guaranteed by math, not prediction. 31 test cases, zero false positives.
session-handoff †	Synthesizes a session into a structured handoff block for clean context-window transfers.

Visual Verification

Skill	What it solves
live-preview †	Mid-build screenshot verification loop. Catches invisible features before they compound. Exits cleanly on non-UI repos. Requires Playwright (optional).
qa †	Browser QA via Playwright — navigates, clicks, fills forms, verifies flows. Web projects only. Gracefully skips if Playwright is not installed.

How skills get added

I find a problem nobody's solved structurally, build and test a skill that solves it, publish it here. The bar is: works on any codebase, zero project-specific assumptions, drops in with a folder copy.

---

Want the full orchestration infrastructure these skills were born from? Citadel — autonomous campaign agents, parallel fleet coordination, lifecycle hooks, and the /do router that dispatches all of it.