asynkron-devtools

asynkron-skills

Developer tool skills for AI coding agents. Works with Claude Code, Codex, Cursor, and Copilot.

General Skills

/cloc

Count lines of code using cloc. Analyze codebase size, language breakdown, and compare code between git refs. Supports 200+ languages and multiple output formats (JSON, YAML, CSV, Markdown).

/quickdup

Detect code duplication using QuickDup. A fast structural clone detector that processes ~100k lines in 500ms. Surfaces duplication candidates for AI-assisted review.

Includes refactoring guidance for 4 types of duplication:

• Structural duplication — same control flow, different injected behavior. Extract the structure, pass behavior in.
• Structural switch duplication — repeated pattern matches on domain types. Often intentional documentation — only refactor at 3+ copies.
• Parameter bundle duplication — same group of parameters passed together. Extract a value object.
• Argument unpacking duplication — repeated defensive parsing of callback arguments. Extract a helper.

/systematic-debug

Two complementary debugging techniques for when the root cause of a bug is unclear.

Test bombs create one small test per hypothesis (H1, H2, H3...). Run them all together — the pass/fail pattern reveals the failing area. Start broad, then add targeted hypotheses based on results.

Layered tests test each stage of a processing pipeline independently (L1, L2, L3...). If L3 fails but L1 and L2 pass, the bug is in stage 3. Works for compilers, HTTP pipelines, ETL, build systems, and any multi-stage architecture.

Use test bombs first to narrow the component, then layered tests to pinpoint the exact stage. Templates for C#, TypeScript, Python, and Go.

/continue-loop

Keep iterating on the next logical project step until a real stop condition is reached. The skill uses a small persisted loop state under $CODEX_HOME/state/continue-loop so agents can choose a bounded next slice, complete it, verify it, record what happened, and either continue or stop cleanly.

/palette

Generate color palettes with Tailwind-scale shades using Asynkron.Palette. A zero-dependency CLI that takes hex colors, a color wheel strategy, and a count, and outputs a complete palette with 11 perceptually uniform shades per color (OKLCH color space). Runs via npx @asynkron/palette (no install required).

Pass one or more base colors (or none for a random palette), pick a strategy (triadic, analogous, complementary, split-complementary, tetradic, evenly-spaced), and set a count. All colors are automatically normalized to matching lightness and chroma so the palette looks cohesive. Mood presets (--vibrant, --pastel) override the target energy level. Use --raw to keep exact input colors.

Colors are named semantically (primary, secondary, tertiary, accent, neutral) with Tailwind shade numbers (50–950). ANSI color output in terminals. Accepts hex with or without # prefix.

/hotmess

Git churn analysis that goes beyond a flat "most-edited files" list. Reads git log --numstat over a window and auto-detects clusters by directory, by filename stem (foo.go + foo_test.go grouped), and by co-change (files modified in the same commit, with Jaccard overlap). Classifies each cluster as hotspot, spec-drift, rewriting, active dev, or stable based on commit count, deletion ratio, test ratio, and co-change overlap, then explains why each cluster is a problem and what to do about it.

Pass a folder, a time window (7 days ago, 1 month ago, ISO date), and optionally an extension or list (.go, .ts,.tsx); auto-detects code files when no extension is given. Add --html /tmp/hotmess.html for a standalone Mermaid module-map visualization with cross-directory co-change edges and collapsible signal tables. Pure Python stdlib + git, no external dependencies.

.NET Skills

/profile

Profile .NET applications using Asynkron.Profiler. A CLI profiler that outputs structured text — no GUI needed. Runs via dnx (no install required).

Five profiling modes: CPU, memory allocations, lock contention, exceptions, and heap snapshots. Can also capture JIT-to-native compilation events to verify inlining.

Includes a full profiling methodology (symptom-to-mode table, reading hot function tables, allocation call graphs, iterative fix-and-measure workflow) and JIT-aware optimization patterns: