Minimal collection of reusable coding skills.
clankermerge— Controlled interactive rebase/merge: auto-resolve obvious conflicts, ask one compact question per real decision.complexitysmash— Audit a codebase for systemic complexity and high-leverage simplifications.cupel— Internal skill for investing research, theses, watchlists, and decision journaling.firehose— Run one project at maximum wall-clock throughput with many parallel agents: frontier planner + cheap workers, a durable ledger, merge on fast gates, fix forward on a slightly-red main, quality enforced at milestone boundaries.ghmonitor— Monitor GitHub Actions runs and fix failures.gavel— Triage open decisions: do the clear and trivial ones directly, close stale ones, and grill the user one at a time, in plain prose with examples and a recommendation, only on calls that need their judgment.groundhog— Mine local Claude/Codex session traces for recurring patterns and friction, then report numbered improvement recommendations (read-only until prompted).localref— Clone or download external references locally for research.moonshot— Strip a feature to first principles, hunt the frontier wherever it lives (research, legendary implementations, other industries), and return ranked, implementable approaches the project has left untapped.orchestrate— Run large multi-phase work as an Opus 5.5 orchestrator (high effort) driving Opus 5.5 subagents and Devin SWE-2 High via devin-harness-run, with a Claude-budget routing rule, a worker-reuse rule of thumb, tight briefs and ground-truth review.performancesmash— Audit systems for real performance bottlenecks and high-leverage optimizations.pilcrow— Internal skill for prose editing, linting, and writing workflows.profilesmash— Orchestrated audit of performance measurement infrastructure (profilers, benchmarks, load tests, tracing, budgets, regression gates): catalog every asset as delete (misleading or dead), unclear, or keep; map which flows are attributable, trustworthy, and regression-guarded across CPU, waiting, latency, memory, stack, round trips, and more; and propose the instruments that are missing.prospect— Benchmark your project against reputable peer projects: clone them locally, extract enforceable practices, audit your own code through their written doctrine, and output independently verified improvements.schemasmash— Orchestrated read-only audit of types, data structures, schema, and serialized shapes for invalid-but-representable states, duplicate sources of truth, and invariants the representation fails to enforce.sounding— Read where a project stands (conversation, milestones, tickets, recent change and trouble, past runs) and return a plan of which skills to run now, which later, and why the rest wait. Flags missing kinds of pass. Advisory, explicit invocation only.spring-cleaning— Do a full codebase cleanup and restructuring pass.testsmash— Orchestrated audit of the whole test and eval approach: catalog every test as delete (frivolous or tautological), unclear, or keep; review expect, property, simulation, e2e, and eval harnesses; and propose new approaches (property, simulation, chaos, soak, formal methods, evals) where real gaps exist.visual-explainer— Turn a topic, plan, codebase, or feature into a single self-contained HTML visual explainer with diagrams, wireframes, and annotated code.wholehog— Implement the best end-state directly with a clean cutover and no legacy leftovers.wholehog-loop— Orchestrate implement-then-score across separate subagents, looping until a target wholehog score is reached or a blocker stops it.wholehog-score— Score how close a current implementation is to the ideal wholehog end-state (higher = closer).
audit-your-codebase— External read-only, agent-orchestrated codebase audit for data structures, state modeling, algorithms, and ownership, by Aaron Francis.impeccable— External skill collection by Paul Bakaus.thermo-nuclear-code-quality-review— External Cursor skill for strict maintainability and code quality reviews.