A curated collection of modular, production-tested Agent Skills for AI coding assistants (Claude Code, Cursor, Antigravity, Codex, OpenCode, and any tool implementing the Agent Skills Specification).
Designed for high-autonomy agents, rigorous engineering standards, verified diffs, and low-slop communication.
Install any or all skills using the open skills CLI:
# Install all skills globally
npx skills add justomat/skills -g
# Install into the current project only
npx skills add justomat/skills
# Install specific skills
npx skills add justomat/skills --skill poteto-mode technical-writing unslop
# List available skills without installing
npx skills add justomat/skills -lWorks out of the box with any tool compliant with the Agent Skills specification, including:
- Claude Code (
~/.claude/skills/or~/.agents/skills/) - Antigravity / agy (
~/.gemini/antigravity-cli/skills/or~/.agents/skills/) - Cursor (
.cursor/skills/or global) - Codex CLI / Copilot (
~/.codex/skills/,~/.copilot/skills/) - OpenCode (
~/.config/opencode/skills/)
Universal principles that sharpen model decision-making before and during implementation.
| Skill | Description |
|---|---|
principle-boundary-discipline |
Separate public contracts from internal mechanics; parse at boundaries. |
principle-build-the-lever |
Build the tool, codemod, or script that performs or proves non-trivial work instead of hand editing. |
principle-encode-lessons-in-structure |
Turn recurring instructions or corrections into lints, types, or runtime checks. |
principle-exhaust-the-design-space |
Prototype competing options side-by-side for novel architecture or UI choices. |
principle-experience-first |
Prioritize end-user delight and polish over implementation convenience. |
principle-fix-root-causes |
Trace symptoms to real roots; avoid nil-check guards that silence bugs. |
principle-foundational-thinking |
Get core types, concurrent actor boundaries, and data structures right first. |
principle-guard-the-context-window |
Offload bulk payloads to subagents; keep summaries in main context. |
principle-laziness-protocol |
Bias toward minimal diffs, zero dead weight, and deleting unnecessary abstractions. |
principle-make-operations-idempotent |
Ensure lifecycle steps converge to the same state across crashes and retries. |
principle-migrate-callers-then-delete-legacy-apis |
Migrate callers and remove deprecated APIs in one atomic pass. |
principle-minimize-reader-load |
Reduce mental hops between questions and answers; collapse single-caller wrappers. |
principle-model-the-domain |
Encode business concepts as explicit types and structures rather than scattered conditionals. |
principle-never-block-on-the-human |
Run reversible actions autonomously and present results for fast human feedback. |
principle-outcome-oriented-execution |
Drive directly toward the target architecture without throwaway transitional code. |
principle-prove-it-works |
Verify against actual artifacts, live outputs, and real runs—never assume "it compiles." |
principle-redesign-from-first-principles |
Integrate new requirements as if they were day-one requirements. |
principle-separate-before-serializing-shared-state |
Eliminate shared state across concurrent actors before introducing locks or queues. |
principle-sequence-verifiable-units |
Break multi-step tasks into self-verifying, reviewable milestone commits and PRs. |
principle-subtract-before-you-add |
Remove obsolete code and dead paths before building new capabilities. |
principle-type-system-discipline |
Make invalid states unrepresentable using strict branded types and exhaustive guards. |
High-leverage workflows for unattended or coordinated agent runs.
| Skill | Description |
|---|---|
poteto-mode |
Complete agent operating style: deliberate subagents, playbooks, unslopped prose, verified work. |
arena |
Run competitive solution branches and pick the optimal implementation. |
automate-me |
Identify repetitive workflows and convert them into automated scripts or skills. |
figure-it-out |
Autonomous problem-solving protocol when facing ambiguity or obscure errors. |
recall |
Reconstruct recent working context, live state, and unresolved tasks before starting work. |
reflect |
Multi-perspective retrospective across transcripts to capture actionable learnings. |
show-me-your-work |
Generate a structured decision log (TSV) documenting every tradeoff and verification step. |
handoff |
Compact active conversation into a structured brief for the next agent session. |
swarm |
Fan out N parallel worker agents to explore or test across large surfaces simultaneously. |
Relentless plan grilling and domain modeling patterns (adapted from Matt Pocock).
| Skill | Description |
|---|---|
grilling |
Grill the user relentlessly about a plan, decision, or architecture tree until every branch is resolved. |
grill-me |
Quick invocation for the grilling interview to stress-test your thinking. |
grill-with-docs |
Relentless interview that concurrently produces ADRs and domain glossaries. |
domain-modeling |
Actively build and sharpen a project's domain glossary (CONTEXT.md) and ADRs. |
Tools for understanding systems, probing history, and crafting robust code.
| Skill | Description |
|---|---|
codebase-design |
Shared vocabulary and principles for designing deep modules with narrow interfaces. |
improve-codebase-architecture |
Scan codebase for architectural friction, shallow modules, and deepening opportunities. |
architect |
Sketch types, interfaces, and module structure before writing implementation code. |
how |
Deep trace of runtime behavior, control flows, and state lifecycles. |
why |
Archeological deep-dive into git history, PRs, issues, and incident rationale. |
interrogate |
Methodically cross-examine requirements and code boundaries to uncover hidden assumptions. |
typescript-best-practices |
Production patterns, strict typing rules, and performance guidelines for TypeScript. |
Ensuring correctness and regression prevention.
| Skill | Description |
|---|---|
tdd |
Test-driven development (red-green-refactor loop) verifying behavior at public seams. |
pstack-tdd |
Targeted test-driven development loop for regression tests and critical logic. |
blast-radius |
Calculate downstream impact and potential breakage across callers before refactoring. |
create-verification-skill |
Author reusable verification scripts and harnesses for complex feature checks. |
maintain-verification-skill |
Keep automated verification suites and checks up to date as systems evolve. |
Direct agent harnesses and integrations.
| Skill | Description |
|---|---|
ego-browser |
Programmatic browser automation, live DOM interaction, and UI verification runtime. |
antigravity-manager |
Guide Claude Code to orchestrate the Antigravity CLI (agy) as an autonomous worker. |
pi-manager |
Direct the pi coding-agent harness with pinned model selection and execution flags. |
biznetgio |
Manage Biznet Gio cloud infrastructure (metal, VMs, storage, IPs) via CLI. |
tiktok-business-messaging |
End-to-end integration with TikTok Business API (DMs, webhooks, auth, media). |
plannotator-compound |
Deep analysis of Plannotator plan archives to improve future prompt and planning accuracy. |
make-bot-ui |
Scaffold webhook-triggered web interfaces to wake bots with JSON payloads. |
setup-pstack |
Configure model assignments per role and workload. |
Ensuring clean, concise, and noise-free documentation and code.
| Skill | Description |
|---|---|
technical-writing |
Diátaxis structure, Google developer style, STE instruction rules, and Global English syntax. |
unslop |
Eliminate AI tropes, buzzwords, filler phrases, and unnecessary hedges. |
teach |
Synthesize how and why into clear, intuitive explanations for humans. |
no-comments |
Write self-explanatory code without noisy, obvious inline comments. |
bro |
Direct, informal, high-signal peer programming tone. |
Every skill in this repository is 100% compliant with the Agent Skills Specification. Validate locally with skills-ref:
# Validate a specific skill
npx skills-ref validate skills/poteto-mode
# Validate all skills
bun run validate
# or
npm run validate