Personal collection of Claude Code skills, reference docs, and configuration for AI-assisted development.
| Path | What it is |
|---|---|
skills/ |
Claude Code skills (SKILL.md + references/), auto-invoked by Claude Code based on their description frontmatter. |
docs/ |
Standalone reference specs (not skills) that codify conventions for a stack. |
config/ |
The Claude Code Context Stack — install scripts and spec for Serena (opt-in) and ponytail, plus the global routing contract. |
agents/ |
Reserved for custom agent definitions. Currently empty. |
| Skill | Use for |
|---|---|
architecture-blueprint |
Domain-first backend/full-stack architecture: modular monoliths, vertical slices, type-driven domain modeling, ADRs. |
gauntlet-loop |
Turn a goal into a paste-ready prompt (or run it as lead) that loops builders and blind critics against a frozen real-world reference until two critics in a row pick ours; runbook plus per-domain bars for code, writing, design, data, research, prompts and detection. Design decisions: skills/gauntlet-loop/README.md. |
security-vuln-gauntlet |
The gauntlet loop applied to authorized vulnerability hunting: a hunter proposes, a blind validator must build a non-destructive PoC before anything counts. |
good-readme |
Write, improve, or audit a README, with a source-grounded drift pass so every documented import, command and flag is checked against the actual manifests and exports. Vendored from adewale/good-readme (MIT); notes: skills/good-readme/README.md. |
macro-analyst |
Structured macro/FX analysis for currency pairs, rate differentials, central bank policy. |
opensrc |
Read the actual source of a dependency (npm/PyPI/crates.io/GitHub) at the installed version via opensrc path, instead of guessing from types or docs. |
rust-bevy-architecture |
Architecture method for Bevy (Rust ECS) game projects — layout, plugins, messages/observers, scheduling, determinism. |
rust-type-driven |
Idiomatic type-driven Rust for domain/backend code — parse-don't-validate, newtypes, illegal states unrepresentable, typed errors, async cancellation discipline. |
rust-wgpu-functional |
Idiomatic Rust for wgpu / bare-metal GPU code, applying functional principles without sacrificing performance. |
sdd-spec |
Spec-Driven Development artifacts — product/domain/feature specs that define WHAT the software must do, with convention detection, cross-referencing instead of duplication, drift detection against code, and a human approval gate. |
worktrunk |
Git worktree management via Worktrunk (wt; git-wt/git wt on Windows) for parallel/isolated work — create, switch, merge, and clean up worktrees by branch name. |
Skills are picked up automatically by Claude Code when their trigger conditions match a request — no manual invocation needed beyond what each SKILL.md describes.
Claude Code only discovers skills from ~/.claude/skills/ (global) and a repo's .claude/skills/ — never from this checkout itself. config/stack-init.sh global deploys the five project-agnostic skills (gauntlet-loop, opensrc, worktrunk, good-readme, sdd-spec) globally; the domain skills deploy per repo with stack-init skills <name>, because a global skill's description is loaded into every session in every project (D48, D58).
docs/RUST_BACKEND_STACK.md— default technology and architectural choices for Rust backend systems.
config/ holds a separate, self-contained setup: two components — Serena (LSP symbol tools via MCP, opt-in per session) and ponytail (minimal-code discipline, default-on) — plus the global routing contract that tells Claude which tool to use for symbols and execution. 3.0 removed graphify, RTK, and Headroom (D51, D52). See config/README.md for quick start and config/claude-code-context-stack.md for the full spec.